One-Year Reserve Risk And Claims Development Results

What One-Year Reserve Risk Measures

A reserve estimate is made at a valuation date, but the next year brings both claim payments and new information about what remains unpaid. One-year reserve risk concerns the difference between the reserve held at the start of the year and the sum of the payments made during the year and the reserve estimated at year-end. It is not simply the difference between the initial reserve and the eventual ultimate claims.

The sign convention matters. This tutorial defines a positive claims development result as favorable: actual payments plus the ending reserve are less than the initial reserve. A negative result is unfavorable. All amounts below are nominal dollars, with no discounting or interest adjustment.

Definition: Let $R_0$ be the reserve for a block of claims at the start of a one-year period, $P_1$ the claims paid during that year, and $R_1$ the reserve for the remaining claims at the end of the year. The one-year claims development result (CDR) is $$ \operatorname{CDR}=R_0-P_1-R_1. $$ Under this convention, a positive CDR is favorable and a negative CDR is adverse. Some sources use the opposite sign; reversing the convention changes the sign of the CDR and its mean, but not its variance or standard deviation.

The ending reserve $R_1$ is a revised estimate, based on information available at the end of the year. It is not the amount ultimately paid in all future years. Consequently, the CDR captures the year’s paid emergence together with the change in the estimate of what remains. Both parts are needed to assess one-year reserve risk.

Conditional Mean and Variance

Let $\mathcal{I}_0$ represent the information available at the initial valuation. If the initial reserve is the conditional expected value of next year’s payments plus the ending reserve, then $R_0=E[P_1+R_1\mid\mathcal{I}_0]$. This is the calibration condition for an unbiased one-year reserve prediction. It does not say that every realized CDR will be zero.

Result: If $R_0=E[P_1+R_1\mid\mathcal{I}_0]$, then $$ E[\operatorname{CDR}\mid\mathcal{I}_0]=0 $$ and $$ \operatorname{Var}(\operatorname{CDR}\mid\mathcal{I}_0) =\operatorname{Var}(P_1+R_1\mid\mathcal{I}_0). $$

The variance expression includes the dependence between payments and the ending reserve. In general, it is not correct to add their variances and omit their covariance. The same principle applies when results from multiple origin years or portfolios are aggregated.

Why the Conditional Results Hold

At the initial valuation, $R_0$ is fixed given $\mathcal{I}_0$. Substituting the CDR definition and applying the reserve calibration gives

E[\operatorname{CDR}\mid\mathcal{I}_0] =R_0-E[P_1+R_1\mid\mathcal{I}_0] =0.

For the variance, subtracting a fixed quantity does not change variance, and multiplying by $-1$ does not change variance either. Therefore,

\begin{aligned} \operatorname{Var}(\operatorname{CDR}\mid\mathcal{I}_0) &=\operatorname{Var}(R_0-(P_1+R_1)\mid\mathcal{I}_0)\\ &=\operatorname{Var}(P_1+R_1\mid\mathcal{I}_0). \end{aligned}

This result is conditional on the initial information and on the stated calibration. In practice, a selected reserve may not equal the model’s conditional expectation, and assumptions may be imperfect. In those circumstances, the expected CDR need not be zero. The variance formula also does not by itself account for uncertainty about the model or selected parameters unless that uncertainty is included in the predictive distribution.

Calculating Risk From a Predictive Distribution

A predictive distribution for $P_1+R_1$ can be translated directly into a distribution for CDR by subtracting each outcome from $R_0$. For discrete outcomes $c_j$ with probabilities $p_j$, calculate the mean as $\sum_j p_jc_j$ and the variance as $\sum_j p_j(c_j-E[\operatorname{CDR}])^2$. When the mean is zero, the variance is the probability-weighted mean of the squared CDRs.

Worked Example: Reserve Risk From Three Possible Outcomes

A hypothetical insurer has an initial reserve of $100,000. Its one-year predictive distribution for payments plus the ending reserve is $60,000 with probability 0.25, $100,000 with probability 0.50, and $140,000 with probability 0.25. Calculate the expected CDR, its variance, and its standard deviation. Use no discounting.

Method. For each outcome, subtract the total of payments and ending reserve from the initial reserve.

\begin{array}{c|c|c} P_1+R_1 & \text{Probability} & \operatorname{CDR}=100{,}000-(P_1+R_1)\\ 60{,}000 & 0.25 & 40{,}000\\ 100{,}000 & 0.50 & 0\\ 140{,}000 & 0.25 & -40{,}000 \end{array}

The expected total is $0.25(60{,}000)+0.50(100{,}000)+0.25(140{,}000)=15{,}000+50{,}000+35{,}000=\$100{,}000$. Thus the expected CDR is zero:

E[\operatorname{CDR}] =0.25(40{,}000)+0.50(0)+0.25(-40{,}000) =\$0.

Because the mean is zero, the variance is the probability-weighted mean of the squared CDRs. The negative result must be squared as a whole:

\begin{aligned} \operatorname{Var}(\operatorname{CDR}) &=0.25(40{,}000^2)+0.50(0^2)+0.25((-40{,}000)^2)\\ &=0.25(1{,}600{,}000{,}000)+0+0.25(1{,}600{,}000{,}000)\\ &=800{,}000{,}000\text{ dollars}^2. \end{aligned}

The standard deviation is $\sqrt{800{,}000{,}000}=\$28{,}284.271\ldots$, or $28,284.27 rounded to the nearest cent. The probabilities sum to 1, the expected total equals the initial reserve, and the expected CDR is zero, as required by the calibration.

Common mistake avoided: Writing $-40{,}000^2$ means $-(40{,}000^2)$, which is negative and is not the square of the CDR. The correct term is $(-40{,}000)^2$.

Reading the Realized Claims Development Result

For a realized one-year result, the calculation uses the initial reserve, actual payments during the year, and the reserve estimate at the end of the year. A payment forecast alone is insufficient: the ending reserve may have changed because of new information about claim severity, settlement timing, or the remaining claims.

Worked Example: Combining Payments With a Revised Ending Reserve

A hypothetical liability portfolio has an initial reserve of $250,000. At the initial valuation, expected payments over the next year are $100,000 and the expected reserve after that year is $150,000. At year-end, actual payments are $120,000 and the revised ending reserve is $145,000. Calculate the realized CDR and interpret its sign.

Method. Use the actual payment amount and the revised ending reserve, not the initial expected values, in the realized CDR.

\begin{aligned} P_1+R_1&=120{,}000+145{,}000=265{,}000,\\ \operatorname{CDR}&=250{,}000-265{,}000=-\$15{,}000. \end{aligned}

The realized CDR is -$15,000, an adverse result under the stated sign convention: payments plus the revised estimate of remaining claims exceed the initial reserve by $15,000. The initial expected components add to the initial reserve, since $100,000+$150,000=$250,000. The realized total is $15,000 above that benchmark.

This does not establish that ultimate claims will be exactly $265,000. The $145,000 is still a reserve estimate for claims not yet paid. The calculation measures the one-year development result using the information available at year-end.

One-Year CDR Is Not Ultimate Prediction Error

Ultimate prediction error compares an initial estimate of ultimate claims with the eventual ultimate amount. The CDR instead compares the initial reserve with one year of payments plus a reserve estimate made after that year. Later payments can still differ from that ending estimate, so the two measures need not be equal or have the same variability.

Worked Example: Separating One-Year Risk From Ultimate Error

Consider a hypothetical two-year claims block with no discounting. First-year payments $P_1$ are $40,000 or $60,000, each with probability 0.50. Independently, second-year payments $P_2$ are $80,000 or $120,000, each with probability 0.50. Calculate the initial reserve, the one-year CDR distribution, and the ultimate prediction error $U-R_0$, where $U=P_1+P_2$.

Initial reserve. The expected first-year payment is $0.50(40{,}000)+0.50(60{,}000)=\$50{,}000$. The expected second-year payment is $0.50(80{,}000)+0.50(120{,}000)=\$100{,}000$. Therefore, $R_0=E[U]=\$150{,}000$.

Ending reserve and CDR. After the first year, the expected remaining payment is $100,000 regardless of the realized $P_1$, because $P_2$ is independent of $P_1$. Thus $R_1=\$100{,}000$. If $P_1=\$40{,}000$, then $\operatorname{CDR}=150{,}000-40{,}000-100{,}000=+\$10{,}000$. If $P_1=\$60{,}000$, then $\operatorname{CDR}=150{,}000-60{,}000-100{,}000=-\$10{,}000$. Each occurs with probability 0.50, so the CDR mean is zero and its variance is

\operatorname{Var}(\operatorname{CDR}) =0.50(10{,}000^2)+0.50((-10{,}000)^2) =100{,}000{,}000\text{ dollars}^2.

Ultimate prediction error. The four equally likely pairs $(P_1,P_2)$ give the following errors:

\begin{array}{c|c|c} P_1 & P_2 & U-R_0\\ 40{,}000 & 80{,}000 & -30{,}000\\ 40{,}000 & 120{,}000 & 10{,}000\\ 60{,}000 & 80{,}000 & -10{,}000\\ 60{,}000 & 120{,}000 & 30{,}000 \end{array}

The mean ultimate error is $(-30{,}000+10{,}000-10{,}000+30{,}000)/4=0$. Its variance is

\begin{aligned} \operatorname{Var}(U-R_0) &=0.25\left((-30{,}000)^2+(10{,}000)^2+(-10{,}000)^2+(30{,}000)^2\right)\\ &=0.25(2{,}000{,}000{,}000)\\ &=500{,}000{,}000\text{ dollars}^2. \end{aligned}

The ultimate error includes uncertainty in both years; the one-year CDR reflects only the first-year result after allowing for the updated expected remaining payment. Their variances differ: $100,000,000 dollars squared for the CDR versus $500,000,000 dollars squared for ultimate error. Common mistake avoided: treating the ending reserve as if it were the actual second-year payment. It is an estimate, so later payment uncertainty remains in ultimate error.

Aggregating Results and Avoiding Common Traps

For several origin years, the portfolio CDR is the sum of their CDRs under the same sign convention. Its variance includes pairwise covariances:

\operatorname{Var}\left(\sum_i \operatorname{CDR}_i\right) =\sum_i\operatorname{Var}(\operatorname{CDR}_i) +2\sum_{i<j}\operatorname{Cov}(\operatorname{CDR}_i,\operatorname{CDR}_j).

The covariance terms matter when origin years share development-factor estimates, calendar-year influences, or other sources of dependence. Assuming independence without justification can understate or overstate portfolio risk. A variance estimate based on a Mack or overdispersed-Poisson framework should therefore be interpreted according to its assumptions and the uncertainty it includes; a variance of process outcomes alone need not represent every source of predictive uncertainty.

Keep the sign convention visible. Here $\operatorname{CDR}=R_0-P_1-R_1$, so an excess of payments plus ending reserve produces a negative result.
Use the revised ending reserve. Do not substitute the initial estimate of remaining claims for $R_1$.
Square negative outcomes correctly. Write $(-c)^2$, not $-c^2$, when calculating a squared deviation.
Do not equate one-year CDR with ultimate error. The CDR uses an updated estimate at the one-year horizon; ultimate error includes subsequent emergence as well.
Check dependence before aggregating variance. The portfolio variance includes covariance terms unless independence is justified.

These distinctions are aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus. In an exam calculation, state the CDR convention, identify whether figures are expected or realized, and verify that the expected result is consistent with the reserve calibration before interpreting its risk measure.

Check Your Understanding

For each problem, use the sign convention $\operatorname{CDR}=R_0-P_1-R_1$. Assume no discounting.

A hypothetical reserve is $75,000. Payments plus ending reserve are $60,000 with probability 0.50, $75,000 with probability 0.20, and $100,000 with probability 0.30. Calculate the expected CDR, variance, and standard deviation. (Answer: Expected CDR $0; variance $300,000,000 dollars squared; standard deviation $17,320.51.)
A portfolio has an initial reserve of $180,000. Actual payments over the year are $72,000, and the revised ending reserve is $116,000. Calculate and interpret the realized CDR. (Answer: -$8,000, adverse.)
A hypothetical initial reserve is $90,000. Payments plus ending reserve are $80,000 with probability 0.20, $90,000 with probability 0.50, and $110,000 with probability 0.30. Calculate the expected CDR. (Answer: -$4,000.)
In a two-year block, $P_1$ is $30,000 or $50,000 with equal probabilities; independently, $P_2$ is $70,000 or $110,000 with equal probabilities. Find the initial reserve, the one-year CDR variance if the ending reserve is $E[P_2]$, and the ultimate error variance. (Answer: Initial reserve $130,000; CDR variance $100,000,000 dollars squared; ultimate error variance $500,000,000 dollars squared.)
Two origin-year CDRs have zero means, variances of 4 and 9 thousand-dollars squared, and covariance of 3 thousand-dollars squared. Calculate the portfolio variance and standard deviation. (Answer: 19 thousand-dollars squared; standard deviation $\sqrt{19}=\$4,358.90$, approximately.)

What You'll Learn