Why a Reinsurance Layer Can Change Capital by More Than Its Expected Recovery
A reinsurance layer changes the distribution of losses retained by an insurer, not just the average amount paid by the reinsurer. The capital effect therefore depends on which parts of the loss distribution are reduced. A layer that responds in the extreme tail may reduce a tail risk measure substantially, even if its expected recovery is modest. Conversely, expected recoveries alone do not show how much capital the layer releases.
This tutorial measures capital as TVaR at a specified confidence level minus expected loss. This is one defined capital basis for making a like-for-like gross-to-net comparison. The result is a measure of capital relief before reinsurance premium, counterparty risk, or other transaction effects; it is not a measure of the layer’s net economic value.
The attachment point is the loss amount the insurer must bear before the layer begins to pay. The limit is the maximum recovery, so the layer stops paying once the loss reaches \(a+L\). For example, a layer attaching at 100 with limit 200 is often described as 200 excess of 100: it pays up to 200 on loss amounts above 100.
For this tutorial, assume the layer is an indemnity contract covering the same aggregate loss \(X\), with full recovery and no reinstatements. The loss and recovery amounts use the same currency and period. No premium, counterparty default, or change in the underlying portfolio is included. Those assumptions isolate the capital effect of the gross and net loss distributions.
Capital Relief: The Exact Gross-to-Net Calculation
For a loss variable \(Z\), define \(\operatorname{VaR}_u(Z)\) as its \(u\)-quantile, using the generalized inverse of its distribution function. TVaR is the average of the quantiles in the upper tail:
This definition applies to both continuous and discrete loss distributions. In a discrete distribution, the tail may cut through the probability mass at the quantile boundary. In that case, include only the fraction of that mass needed to make the upper tail probability equal to \(1-\alpha\). Simply averaging all outcomes at or above the VaR can overstate or understate TVaR.
The key capital-relief identity follows directly from the definition of retained loss and linearity of expectation:
The last equality holds because \(N_{a,L}=X-R_{a,L}\), so \(E[N_{a,L}]=E[X]-E[R_{a,L}]\). The identity separates two effects: the reduction in TVaR, less the expected recovery. Expected recovery is subtracted because this capital measure is already net of expected loss. The identity is exact, but it does not make TVaR additive: the net distribution must still be calculated to find \(\operatorname{TVaR}_\alpha(N_{a,L})\).
Worked Calculations
Worked Example: Calculating Relief for a Single Layer
A hypothetical insurer models one-period aggregate loss \(X\), in thousands of dollars, as follows: \(X=0\) with probability 0.50, \(X=100\) with probability 0.30, and \(X=300\) with probability 0.20. It considers a layer attaching at 100 with limit 100. Calculate capital relief at \(\alpha=0.80\), using \(C_\alpha(Z)=\operatorname{TVaR}_\alpha(Z)-E[Z]\). Assume full recovery and no premium or counterparty risk.
Method. Calculate the layer recovery and retained loss in every state. Then calculate expected losses and the upper 20% tail averages for both distributions.
Step 1: Gross expected loss.
Step 2: Gross TVaR. The largest 20% of probability consists exactly of the \(X=300\) outcome, which has probability 0.20. Thus \(\operatorname{TVaR}_{0.80}(X)=300\), and gross capital is \(300-90=210\).
Step 3: Recovery and net loss. The layer pays nothing at \(X=0\) or \(X=100\), and pays \(\min(300-100,100)=100\) at \(X=300\). Therefore the net losses are 0, 100, and 200 in the three states. Expected recovery is \((100)(0.20)=20\), and expected net loss is \(90-20=70\).
Step 4: Net TVaR and capital relief. The upper 20% of net loss is the \(N=200\) outcome, so net TVaR is 200 and net capital is \(200-70=130\). Capital relief is \(210-130=80\) thousand dollars.
Check. The TVaR reduction is \(300-200=100\). Subtracting expected recovery gives \(100-20=80\), matching the direct capital comparison. The common mistake of reporting 100 as capital relief would confuse TVaR reduction with capital relief.
Worked Example: Comparing Attachment Points and Limits
Use the hypothetical gross loss distribution from the previous example and the same \(\alpha=0.80\) capital definition. Compare three fully collectible layers: (A) attachment 0, limit 100; (B) attachment 100, limit 100; and (C) attachment 100, limit 200. All losses are in thousands of dollars. Determine each layer’s capital relief.
Gross benchmark. From the distribution, \(E[X]=90\), \(\operatorname{TVaR}_{0.80}(X)=300\), and gross capital is \(300-90=210\).
Layer A: attachment 0, limit 100. Recoveries at gross losses \(0,100,300\) are \(0,100,100\). Net losses are \(0,0,200\). Expected recovery is \((0.30+0.20)(100)=50\), so expected net loss is \(90-50=40\). The upper 20% of net loss is 200, giving net capital \(200-40=160\). Relief is \(210-160=50\).
Layer B: attachment 100, limit 100. Recoveries are \(0,0,100\), and net losses are \(0,100,200\). Expected recovery is \((0.20)(100)=20\); expected net loss is \(70\). The upper 20% net loss is 200, so net capital is \(200-70=130\). Relief is \(210-130=80\).
Layer C: attachment 100, limit 200. Recoveries are \(0,0,200\), and net losses are \(0,100,100\). Expected recovery is \((0.20)(200)=40\); expected net loss is \(50\). The net loss is 100 in the entire upper 20%, so net TVaR is 100 and net capital is \(100-50=50\). Relief is \(210-50=160\).
Comparison and check. The three relief amounts are 50, 80, and 160 thousand dollars. For each layer, the identity gives the same result: gross-to-net TVaR reduction less expected recovery. For example, Layer C reduces TVaR by \(300-100=200\), and \(200-40=160\). A larger limit need not produce relief in direct proportion to the limit; its effect depends on how the layer changes the losses in the relevant tail.
Worked Example: Measuring the Incremental Relief From Widening a Layer
A hypothetical one-period aggregate loss \(X\), in thousands of dollars, has this distribution: 0 with probability 0.40, 100 with probability 0.30, 200 with probability 0.20, and 500 with probability 0.10. At \(\alpha=0.75\), compare a layer attaching at 100 with limit 100 against the same layer widened to limit 200. Calculate the capital relief from the extra limit, over and above the first layer.
Step 1: Gross capital. Expected loss is
The upper 25% consists of all 10% at loss 500 and 15% out of the 20% at loss 200. Therefore \(\operatorname{TVaR}_{0.75}(X)=((0.10)(500)+(0.15)(200))/0.25=(50+30)/0.25=320\). Gross capital is \(320-120=200\).
Step 2: Capital with limit 100. Recoveries at gross losses \(0,100,200,500\) are \(0,0,100,100\). Net losses are \(0,100,100,400\), with probabilities \(0.40,0.30,0.20,0.10\). Expected recovery is \(0+0+20+10=30\), so expected net loss is 90. The upper 25% of net loss contains all 10% at 400 and 15% at 100. Net TVaR is \(((0.10)(400)+(0.15)(100))/0.25=55/0.25=220\). Net capital is \(220-90=130\), so relief from this layer is \(200-130=70\).
Step 3: Capital with limit 200. Recoveries are now \(0,0,100,200\), and net losses are \(0,100,100,300\). Expected recovery is \(0+0+20+20=40\), and expected net loss is \(120-40=80\). Net TVaR is \(((0.10)(300)+(0.15)(100))/0.25=45/0.25=180\). Net capital is \(180-80=100\), so total relief is \(200-100=100\).
Step 4: Isolate the incremental effect. Widening the limit from 100 to 200 increases total relief by \(100-70=30\) thousand dollars. The extra expected recovery is \(40-30=10\), and net TVaR falls by \(220-180=40\). Thus the incremental capital relief is \(40-10=30\), consistent with the difference in total capital relief.
Check. Both calculations use the same gross benchmark and confidence level. The extra layer band pays only when \(X=500\), but its effect on the upper-tail average is measured using the resulting net distribution, not by simply multiplying its limit by the probability of a claim.
Interpreting Marginal Relief and Avoiding Traps
For two layer designs on the same portfolio and capital basis, compare their net capital amounts or their capital relief amounts against the same gross benchmark. To isolate an incremental band, subtract the capital relief of the narrower arrangement from that of the wider arrangement. Equivalently, compare the capital of the two resulting net distributions. This finite-difference comparison is useful even when the loss distribution is discrete and a smooth derivative with respect to attachment or limit would be inappropriate.
The calculation also makes clear why the word “marginal” requires care. An incremental layer band has a recovery distribution of its own, but its capital effect depends on the full net distribution after adding that band. In particular, do not assume that the band’s capital relief equals its expected recovery, its limit, or the decrease in TVaR alone.
- Wrong: Use the fall in TVaR as capital relief. Correction: For capital defined as TVaR minus mean, subtract expected recovery from the fall in TVaR.
- Wrong: Average every loss at or above the VaR in a discrete distribution. Correction: Include only the probability mass required to form the upper tail of size \(1-\alpha\).
- Wrong: Compare gross and net capital at different confidence levels or on different loss bases. Correction: Hold the capital definition and comparison basis constant.
- Wrong: Treat calculated capital relief as net value to the insurer. Correction: This calculation excludes premium, counterparty risk, and other effects; it isolates only the specified gross-to-net capital change.
Check Your Understanding
For each problem, use capital \(C_\alpha(Z)=\operatorname{TVaR}_\alpha(Z)-E[Z]\), assume full recovery, and ignore premium and counterparty risk. Losses and answers are in thousands of dollars.
- A loss is 0 with probability 0.60, 80 with probability 0.25, and 240 with probability 0.15. At \(\alpha=0.85\), calculate capital relief for a layer attaching at 80 with limit 80. (Answer: 68)
- A loss is 0 with probability 0.50, 100 with probability 0.30, and 300 with probability 0.20. At \(\alpha=0.80\), compare a layer attaching at 0 with limit 100 and a layer attaching at 100 with limit 100. Give the relief for each, in that order. (Answer: 50; 80)
- A loss is 0 with probability 0.40, 100 with probability 0.30, 200 with probability 0.20, and 500 with probability 0.10. At \(\alpha=0.75\), calculate relief for a layer attaching at 100 with limit 100. (Answer: 70)
- Use the distribution and confidence level in Question 3. Calculate the relief from a layer attaching at 200 with limit 100. (Answer: 30)