Credibility-Weighted Ultimate Loss Estimates

Why Weight Ultimate Estimates?

An accident period can have an observed ultimate loss estimate that reflects its own experience and an external expected-loss estimate that supplies a broader reference point. A credibility-weighted estimate combines the two: experience receives more weight when the credibility factor is higher, and the external estimate receives more weight when it is lower. The result is an estimated ultimate loss, not an unpaid reserve. Paid losses must still be deducted to obtain the reserve.

This tutorial treats the credibility factors as given inputs. It focuses on applying them consistently by accident period and reconciling the resulting reserves. All amounts in the examples are hypothetical and are stated in thousands of dollars. Assume estimates and paid losses are measured at the same valuation date, on the same covered-loss and gross-or-net basis, and without discounting future payments.

Definition: For accident period $i$, let $U_i^{\mathrm{obs}}$ be the observed ultimate loss estimate, $U_i^{0}$ the external expected ultimate loss estimate, $Z_i$ the credibility factor, and $P_i$ the paid loss to date. With $0 \leq Z_i \leq 1$, the credibility-weighted ultimate estimate and its unpaid reserve estimate are $$ \widehat{U}_i = Z_i U_i^{\mathrm{obs}} + (1-Z_i)U_i^{0}, \qquad \widehat{R}_i = \widehat{U}_i-P_i. $$

The two ultimate estimates must refer to the same accident period and be on a comparable basis. For example, a gross observed ultimate should not be blended with a net-of-reinsurance external estimate unless the intended calculation explicitly reconciles those bases first. Paid loss is subtracted after the ultimate estimates have been blended because both inputs to the blend are ultimate amounts.

From a Credibility Blend to an Unpaid Reserve

The calculation is performed period by period. First, multiply the observed ultimate by its credibility factor. Then multiply the external expected ultimate by the complementary weight $1-Z_i$. Add the weighted amounts to obtain the blended ultimate. Finally, subtract the period's paid loss. The period-specific results can then be summed.

Check the basis.
Confirm that observed ultimate, external ultimate, and paid loss are for the same period, valuation date, and loss basis.

Blend ultimate estimates.
Calculate $Z_i U_i^{\mathrm{obs}}+(1-Z_i)U_i^{0}$. The two weights sum to one.

Calculate unpaid reserve.
Subtract paid loss once: $\widehat{R}_i=\widehat{U}_i-P_i$.

Reconcile totals.
Check that the sum of period reserves equals the sum of blended ultimates minus the sum of paid losses.

Worked Example: Two Accident Periods and a Total Reserve Check

A hypothetical insurer has the following estimates, all in thousands of dollars. Period 1 has observed ultimate 1,200, external ultimate 1,000, credibility 0.60, and paid loss 800. Period 2 has observed ultimate 700, external ultimate 900, credibility 0.30, and paid loss 500. Calculate each blended ultimate, each reserve, and the total reserve.

Method. Blend the two ultimate estimates within each period, then subtract that period's paid loss.

Period 1. The external weight is $1-0.60=0.40$. The weighted observed estimate is $0.60(1{,}200)=720$, and the weighted external estimate is $0.40(1{,}000)=400$. Thus

\widehat{U}_1=720+400=1{,}120, \qquad \widehat{R}_1=1{,}120-800=320.

Period 2. The external weight is $1-0.30=0.70$. The weighted observed estimate is $0.30(700)=210$, and the weighted external estimate is $0.70(900)=630$. Thus

\widehat{U}_2=210+630=840, \qquad \widehat{R}_2=840-500=340.

The total reserve is $320+340=660$ thousand dollars. Check by aggregating the blended ultimates and paid losses: total blended ultimate is $1{,}120+840=1{,}960$, and total paid is $800+500=1{,}300$. Therefore $1{,}960-1{,}300=660$ thousand dollars, matching the sum of period reserves.

Common mistake avoided. Do not substitute unverified reserve amounts for the period calculations. In particular, the first period's reserve is $1{,}120-800=320$, not 360, and the second is $840-500=340$, not 360. The aggregate check confirms the corrected total of 660.

Why the Aggregate Check Works

The total-reserve reconciliation follows directly from the definition of each period's reserve. It is a useful arithmetic check, but it does not mean that the period-specific credibility factors can be discarded. Each period must first receive its own stated credibility weight.

Result: For $n$ accident periods, if each period's reserve is calculated as its blended ultimate less its paid loss, then $$ \sum_{i=1}^{n}\widehat{R}_i = \sum_{i=1}^{n}\widehat{U}_i-\sum_{i=1}^{n}P_i = \sum_{i=1}^{n}\left[Z_iU_i^{\mathrm{obs}}+(1-Z_i)U_i^{0}\right]-\sum_{i=1}^{n}P_i. $$

Justification. By definition, $\widehat{R}_i=\widehat{U}_i-P_i$. Summing this equality over periods gives $\sum_i\widehat{R}_i=\sum_i(\widehat{U}_i-P_i)$. Distributing the sum gives $\sum_i\widehat{U}_i-\sum_iP_i$. Substituting the period-level blend for each $\widehat{U}_i$ gives the displayed result. No assumption that the $Z_i$ are equal is needed.

Worked Example: Unequal Credibility Factors Across Periods

A hypothetical specialty portfolio has two accident periods, with all amounts in thousands of dollars. Period A has observed ultimate 1,500, external ultimate 1,200, credibility 0.80, and paid loss 1,100. Period B has observed ultimate 900, external ultimate 750, credibility 0.20, and paid loss 600. Calculate the period reserves and total. Also assess the result of replacing the two factors with their simple average.

Period A. The external weight is $0.20$. The weighted estimates are $0.80(1{,}500)=1{,}200$ and $0.20(1{,}200)=240$. Hence the blended ultimate is $1{,}440$, and the reserve is $1{,}440-1{,}100=340$.

Period B. The external weight is $0.80$. The weighted estimates are $0.20(900)=180$ and $0.80(750)=600$. Hence the blended ultimate is $780$, and the reserve is $780-600=180$.

The total reserve is $340+180=520$. As a check, the total blended ultimate is $1{,}440+780=2{,}220$, and total paid is $1{,}100+600=1{,}700$. Their difference is $2{,}220-1{,}700=520$.

Why a simple average fails. The simple average credibility is $(0.80+0.20)/2=0.50$. Applying it to the aggregate observed and external ultimates would give $0.50(2{,}400)+0.50(1{,}950)=2{,}175$, and an apparent reserve of $2{,}175-1{,}700=475$. This differs from the period-level result of 520 because the periods have different credibility factors and different observed-to-external differences. The correct method retains each period's factor.

Common mistake avoided. A simple average of period factors is not generally a valid factor for aggregate ultimates. Calculate each period's blend first, then add the resulting amounts.

Interpretation and Practical Checks

A credibility factor of 1 assigns all weight to the observed ultimate, while a factor of 0 assigns all weight to the external ultimate. For a factor between 0 and 1, the blended ultimate lies between the two input ultimates: it is a weighted average with nonnegative weights that sum to one. This provides a quick reasonableness check before the paid losses are deducted.

The reserve need not lie between the two ultimate estimates because it is the blended ultimate less paid loss. The essential check is that it equals the estimated ultimate less the appropriate paid amount. Do not floor a negative result at zero unless the problem explicitly specifies such a rule; a floor would change the stated calculation.

A useful exam discipline is to label each value before using it: observed ultimate, external ultimate, credibility, paid loss, blended ultimate, or reserve. This helps prevent a basis error, such as treating an external ultimate as if it were an external unpaid estimate. If an external input is explicitly an unpaid estimate rather than an ultimate estimate, it must first be put on a comparable basis before applying the ultimate-blending formula.

Key takeaway: Credibility-weight the ultimate estimates on a consistent basis, subtract paid loss once for each accident period, and reconcile the sum of period reserves to total blended ultimate less total paid. Keep period-specific credibility factors unless a valid aggregate weighting is established.

Check Your Understanding

For each problem, calculate the requested estimate using the stated period-specific credibility factors. Amounts are in thousands of dollars unless otherwise stated.

An accident period has observed ultimate 640, external ultimate 700, credibility 0.25, and paid loss 500. Find the blended ultimate and unpaid reserve. (Answer: blended ultimate 685; reserve 185)
Period 1 has observed ultimate 900, external ultimate 800, credibility 0.50, and paid loss 620. Period 2 has observed ultimate 500, external ultimate 650, credibility 0.20, and paid loss 410. Find each reserve and the total reserve. (Answer: reserves 230 and 210; total 440)
Three periods have, respectively, observed ultimates 1,100, 800, and 600; external ultimates 1,000, 900, and 700; credibility factors 0.70, 0.40, and 0.25; and paid losses 750, 600, and 500. Find the total reserve and verify it by aggregate blended ultimate less aggregate paid. (Answer: period reserves 320, 260, and 175; total 755; aggregate check $2{,}605-1{,}850=755$)
Period A has observed ultimate 2,400, external ultimate 2,000, credibility 0.50, and paid loss 1,600. Period B has observed ultimate 1,500, external ultimate 1,800, credibility 0.20, and paid loss 1,100. Find the total reserve. (Answer: period reserves 600 and 640; total 1,240)
Period 1 has observed ultimate 500, external ultimate 700, credibility 1.00, and paid loss 430. Period 2 has observed ultimate 800, external ultimate 700, credibility 0.00, and paid loss 610. Find both reserves and explain which ultimate estimate determines each blend. (Answer: reserves 70 and 90; Period 1 uses the observed ultimate, and Period 2 uses the external ultimate)

What You'll Learn