Expected Ultimate Losses Under Chain Ladder Assumptions

Why a Chain Ladder Projection Can Be an Expected Ultimate

The chain ladder method projects a cumulative loss amount forward using selected development factors. Its usual calculation gives a point estimate, but it can also be interpreted as a conditional expected ultimate loss when the development factors represent the appropriate conditional means. That interpretation is useful: it says what the projection estimates, while making clear that the projection is not a guarantee of the amount that will ultimately emerge.

This tutorial focuses on that interpretation and its assumptions. It does not require a particular distribution for future losses. The essential condition is about conditional expectations: given the information available at a development age, the next cumulative amount has an expected value equal to the current amount multiplied by the applicable factor.

Definition: Let $C_{i,j}$ be the cumulative loss for origin year $i$ at development age $j$, and let $U_i$ be its ultimate cumulative loss. Let $f_j$ be the selected factor from age $j$ to age $j+1$. The chain ladder conditional mean assumption is $$ E[C_{i,j+1}\mid \mathcal{F}_{i,j}] = f_j C_{i,j}, $$ where $\mathcal{F}_{i,j}$ represents the information available for that origin year through age $j$. If there is a tail factor $t$, it represents the final development from the last modeled age to ultimate.

A factor of 1.25, for example, says that the conditional expected cumulative amount at the next age is 1.25 times the current cumulative amount. The assumption concerns a conditional mean, not a promise that every origin year will develop by exactly 25%. Future amounts may be above or below their conditional expectations.

For the interpretation below, assume the selected factors are fixed, finite, and equal to the relevant conditional means. Assume the cumulative amounts have finite expectations and that the modeled development reaches ultimate, either at a final age or through an included tail factor. No interest or mortality assumption is involved: this is an undiscounted loss-development calculation. If factors are estimated from data rather than known, the ordinary chain ladder projection is generally a plug-in estimate using those factors, not an exact expectation that includes uncertainty in the factor selections.

Deriving the Expected Ultimate Formula

Suppose an origin year is observed at age $k$, and the modeled future development runs through age $n$. If all future transitions satisfy the conditional mean assumption, then the expected ultimate given the information at age $k$ is the current cumulative amount multiplied by every remaining age-to-age factor and the tail factor, if applicable.

E[U_i\mid\mathcal{F}_{i,k}] = C_{i,k}\left(\prod_{j=k}^{n-1} f_j\right)t

The reason is repeated application of the law of iterated expectations. For two future steps,

\begin{aligned} E[C_{i,k+2}\mid\mathcal{F}_{i,k}] &= E\!\left[E[C_{i,k+2}\mid\mathcal{F}_{i,k+1}]\mid\mathcal{F}_{i,k}\right]\\ &= E[f_{k+1}C_{i,k+1}\mid\mathcal{F}_{i,k}]\\ &= f_{k+1}E[C_{i,k+1}\mid\mathcal{F}_{i,k}]\\ &= f_{k+1}f_kC_{i,k}. \end{aligned}

The second equality uses the conditional mean assumption at age $k+1$. The third uses that $f_{k+1}$ is fixed, and the last uses the assumption at age $k$. Repeating this argument gives the product of all remaining factors. A fixed tail factor is included in the same way if its conditional mean relationship is assumed. Independence of future development steps is not required for this expectation calculation; the stated conditional mean relationships are what matter.

Worked Example: Projecting a Paid Cumulative Loss to Ultimate

A hypothetical commercial property portfolio has a paid cumulative loss of $12.000 million at age 24 months. Selected factors are 1.50 from 24 to 36 months, 1.20 from 36 to 48 months, 1.05 from 48 to 60 months, and a 1.02 tail factor from 60 months to ultimate. Treat these fixed factors as the conditional mean factors in the model. Calculate the conditional expected ultimate and expected future development beyond paid to date.

Method and assumptions. Apply the chain ladder expected ultimate formula. The factors are fixed and each is the conditional mean multiplier for its transition. There is no interest or mortality assumption, and no parametric distribution is specified; only the conditional means are used.

First project from 24 to 36 months:

12.000(1.50)=18.000\text{ million}.

Then project from 36 to 48 months:

18.000(1.20)=21.600\text{ million}.

Then project from 48 to 60 months:

21.600(1.05)=22.680\text{ million}.

Finally apply the tail factor:

22.680(1.02)=23.1336\text{ million}.

Thus the conditional expected ultimate is $23.1336 million. Expected future development beyond the current paid amount is

23.1336-12.000=11.1336\text{ million}.

Check. Multiplying the factors together gives $1.50(1.20)(1.05)(1.02)=1.9278$, and $12.000(1.9278)=23.1336$ million. The result agrees with the sequential projection.

Common mistake avoided: The factors are multiplied, not added to the starting loss. Adding the factors to $12 million would not represent successive percentage development.

What the Expected Ultimate Does—and Does Not—Mean

A conditional expectation is an average over possible future outcomes, given the information used in the conditioning. The actual ultimate for one origin year can differ from the projected expected ultimate. The chain ladder product does not, by itself, specify the variance of that ultimate, the probability that it will exceed a chosen amount, or a prediction interval.

There are also two different uses of the word “expected” that should not be confused. If the true conditional mean factors are known and the assumptions hold, the product is a conditional expectation. In practice, actuaries select factors from observed development data and apply those selections to a current cumulative amount. Unless factor uncertainty is included in a separate model, that calculation is a plug-in estimate of the conditional expected ultimate. It is not automatically a conditional expectation that averages over uncertainty in the selected factors.

Worked Example: Incurred Ultimate and Unpaid Loss Are Different Quantities

A hypothetical liability book has incurred cumulative losses of $8.400 million at age 36 months. The selected conditional mean factors are 1.25 from 36 to 48 months, 1.08 from 48 to 60 months, and a 1.03 tail factor. Paid losses to date are $5.100 million. Calculate the expected ultimate, the amount above incurred to date, and the expected unpaid loss measured from paid to date.

Method and assumptions. Project the incurred cumulative amount through the remaining factors. Treat the factors as fixed conditional mean factors; no interest rate, mortality assumption, or parametric distribution is used.

At 48 months, the projected incurred cumulative amount is

8.400(1.25)=10.500\text{ million}.

At 60 months, it is

10.500(1.08)=11.340\text{ million}.

After applying the tail factor, the conditional expected ultimate is

11.340(1.03)=11.6802\text{ million}.

The amount above incurred to date is

11.6802-8.400=3.2802\text{ million}.

But expected unpaid loss measured from paid losses is ultimate minus paid to date:

11.6802-5.100=6.5802\text{ million}.

Check. The $3.2802 million difference is the expected increase from current incurred to ultimate; the $6.5802 million difference is the expected amount still unpaid from current paid losses. They differ by $3.3000 million, which is the current difference between incurred and paid: $8.400-5.100=3.300$.

Common mistake avoided: Subtracting incurred losses from ultimate does not give total unpaid loss when some claims are already incurred but not yet paid. For a paid-loss reserve, subtract paid losses to date.

Conditional Mean Factors Versus Development Ratios

A development ratio observed in one origin year is a realized ratio of cumulative amounts. A selected factor is a value chosen to represent development at that age. The conditional mean assumption is stronger and more precise than saying that observed ratios have looked stable: it says that, conditional on the information at age $j$, the next cumulative amount has mean $f_jC_{i,j}$. The factors need not describe every observed origin year exactly.

The formula also relies on using the factors in the correct order and from the correct starting age. An origin year observed at age 36 months uses factors beginning at 36 months, not the factor from the earliest development age. A tail factor is included only if it represents development beyond the final modeled age and has not already been incorporated into another selected factor.

Worked Example: A Discrete Development Model

A hypothetical specialty insurer has a paid cumulative amount of $2.000 million at a particular age. For the next transition, the cumulative amount is multiplied by 1.20 with probability 0.75 and by 1.60 with probability 0.25. For the following transition, it is multiplied by 1.10 with probability 1.00. Assume these are the complete conditional distributions for the two transitions, with the second transition’s multiplier fixed at 1.10 regardless of the first outcome. Find the conditional expected ultimate after the two transitions and compare it with the product of the conditional mean factors.

Method and assumptions. Use the stated discrete distributions to calculate the conditional mean multipliers, then apply the chain ladder conditional expectation formula. There is no interest or mortality assumption; the distributions and parameters are given explicitly.

The first transition’s conditional mean factor is

$$ (0.75)(1.20)+(0.25)(1.60)=0.90+0.40=1.30. $$

The second transition’s factor is 1.10, since it occurs with probability 1.00. Therefore the product-of-means projection is

2.000(1.30)(1.10)=2.860\text{ million}.

Check directly by enumerating the outcomes. With probability 0.75, the first amount is $2.000(1.20)=2.400$ million, followed by $2.400(1.10)=2.640$ million. With probability 0.25, the first amount is $2.000(1.60)=3.200$ million, followed by $3.200(1.10)=3.520$ million. The expected ultimate is

(0.75)(2.640)+(0.25)(3.520)=1.980+0.880=2.860\text{ million}.

Check. Direct enumeration agrees with the product of conditional mean factors. The example’s probabilities describe a hypothetical loss-development model, not observed industry experience.

Common mistake avoided: Do not treat one possible realized path, such as two multipliers both equal to their largest values, as the expected ultimate. The expectation weights each possible outcome by its probability.

Assumptions and Practical Limits

The expected-ultimate interpretation is most direct when the following conditions are reasonable:

Appropriate conditional means: Each factor represents the expected next cumulative amount relative to the current cumulative amount, conditional on the information available at that age.
Factors treated as fixed: The derivation uses fixed factors. If factors are estimated, treating them as fixed produces a plug-in projection and does not account for estimation uncertainty.
Consistent measurement: The cumulative amounts and factors use consistent definitions, such as the same loss basis and development ages. Mixing bases can invalidate the conditional mean relationship.
Represented future development: All remaining development is covered by the selected factors and any tail factor. Omitting material development means the calculation is not an expected ultimate under the stated model.
Finite expectations: The conditional expectations used in the derivation must exist.

If any of these conditions is doubtful, the chain ladder product can remain a useful projection, but its interpretation as a conditional expected ultimate requires qualification. In particular, using a factor does not establish that its conditional mean assumption is true. Assessing whether development patterns support that assumption requires examination of the data and the selected factors.

Key takeaway: Under fixed conditional mean factors, repeated application of the law of iterated expectations gives a conditional expected ultimate equal to the current cumulative loss multiplied by all remaining factors. With factors estimated from data, the usual result is a plug-in estimate; it does not automatically include factor uncertainty or describe the distribution of ultimate losses.

Check Your Understanding

For each problem, use the stated factors and assumptions. Amounts are hypothetical and undiscounted.

A paid cumulative amount is $5.000 million at the valuation age. The remaining conditional mean factors are 1.40 and 1.10. Find the expected ultimate. (Answer: $7.700 million)
An incurred cumulative amount is $3.200 million. The remaining factors are 1.25, 1.12, and a 1.05 tail factor. Find the expected ultimate. (Answer: $4.704 million)
A paid cumulative amount is $6.000 million. Apply factors 1.30, 1.08, and a 1.02 tail factor. Find the expected ultimate and expected future development beyond paid to date. (Answer: Expected ultimate $8.59248 million; future development $2.59248 million)
Given the information at the current age, the conditional expected cumulative amount after one transition is 1.20 times the current $4.000 million, and the next conditional mean factor is 1.25. Find the conditional expected ultimate after both transitions. (Answer: $6.000 million)
A hypothetical book has incurred losses of $4.100 million and paid losses of $2.400 million. Its remaining incurred development factors are 1.10 and 1.05. Find the expected ultimate and the expected unpaid amount measured from paid losses. (Answer: Expected ultimate $4.7355 million; expected unpaid loss $2.3355 million)

What You'll Learn