What Basis Risk Measures
An insurance-linked payment may depend on an index, an external loss estimate, or another trigger rather than on the insurer’s own claim amount. The payment can therefore differ from the loss that needs financing. Basis risk is the risk that this mismatch leaves the insurer with less protection, or a different pattern of protection, than intended.
For example, a hypothetical insurer might receive a payment when a regional wind index exceeds a threshold. Its own portfolio loss depends not only on the regional wind reading but also on where insured properties are located, their construction, and the claims that arise. A high index payment can coincide with a moderate portfolio loss; a severe portfolio loss can occur when the index does not trigger enough payment.
The payout function \(g\) is part of the contract. It may include an attachment point, a payout rate, a cap, or other specified terms. For measurement, the important input is the joint behavior of \(X\) and \(Z\): which actual losses occur with which index values, and how often. Knowing only the separate distributions of \(X\) and \(Z\) does not determine the mismatch.
Sources of Mismatch
Several features can separate a portfolio’s actual loss from an external trigger. One event can create more than one source of basis risk.
- Geographic mismatch: the index measures conditions over a broad region, while the insured exposure is concentrated in a smaller area. Local conditions can differ from the regional measure.
- Exposure mismatch: an index may represent a class of assets or an industry total, while the insurer’s portfolio has a different mix of locations, policy limits, deductibles, or insured values.
- Vulnerability mismatch: losses depend on how the portfolio responds to an event. Two portfolios exposed to similar measured conditions can have different damage because of differences in construction, occupancy, or mitigation.
- Timing mismatch: the index’s observation period may not line up with the period in which claims arise or are counted. A short measurement window can miss conditions relevant to the insurer’s loss period.
- Measurement and reporting mismatch: an external index may use estimates, delayed reports, or a different definition of loss from the insurer’s own claims records.
- Trigger-shape mismatch: a threshold, payout rate, or cap may not rise in step with actual losses. A payment can begin too late, grow too slowly, or stop increasing while losses continue to rise.
These sources are useful diagnostics, but the ultimate assessment is quantitative: evaluate actual losses and contractual payments together across plausible joint outcomes. Correlation can summarize part of the relationship between two variables, but it does not by itself describe how often a trigger pays too little or too much in the scenarios that matter most.
Scenario-Based Measurement
A practical method is to set out mutually exclusive joint scenarios for actual loss \(X\) and index \(Z\), assign each scenario a probability, and apply the contract’s payout function to its index value. All scenario probabilities should sum to one. Work in a consistent unit, such as dollars or thousands of dollars.
Useful summaries include expected uncovered loss \(E[U]\), expected excess recovery \(E[Q]\), and the probability that uncovered loss exceeds a specified amount \(c\), \(P(U>c)\). The full scenario distribution can also show whether protection is most deficient in severe-loss outcomes. Expected recovery alone is not enough: it does not show whether payments coincide with losses.
Worked Example: Expected Uncovered Loss Under an Index Payout
A hypothetical insurer considers an index-based cover for one year. Amounts are in thousands of dollars. The specified payment is \(R=\min(800,\max(0,20(Z-40)))\), where \(Z\) is the index reading. The joint scenarios below are exhaustive and mutually exclusive. Calculate the payment, uncovered loss, and excess recovery in each scenario, then find their expectations.
| Probability | Actual loss \(X\) | Index \(Z\) |
|---|---|---|
| 0.20 | 200 | 20 |
| 0.30 | 600 | 60 |
| 0.30 | 1,000 | 50 |
| 0.20 | 400 | 90 |
Method. Apply the specified payout to \(Z\) first. Then calculate \(U=(X-R)^+\) and \(Q=(R-X)^+\) separately. The payout is zero below an index value of 40, increases by 20 thousand dollars per index point above 40, and is capped at 800 thousand dollars.
Scenario 1: \(Z=20\), so \(R=\min(800,\max(0,20(20-40)))=0\). Thus \(U=(200-0)^+=200\) and \(Q=(0-200)^+=0\).
Scenario 2: \(Z=60\), so \(R=\min(800,20(20))=400\). Thus \(U=(600-400)^+=200\) and \(Q=0\).
Scenario 3: \(Z=50\), so \(R=\min(800,20(10))=200\). Thus \(U=(1{,}000-200)^+=800\) and \(Q=0\).
Scenario 4: \(Z=90\), so the uncapped payment is \(20(50)=1{,}000\), but the cap makes \(R=800\). Thus \(U=(400-800)^+=0\) and \(Q=(800-400)^+=400\).
Weight each scenario result by its probability:
The expected uncovered loss is $340,000, and the expected excess recovery is $80,000. As a check, \(E[X]=0.20(200)+0.30(600)+0.30(1{,}000)+0.20(400)=600\), while \(E[R]=0.20(0)+0.30(400)+0.30(200)+0.20(800)=340\), all in thousands. Therefore \(E[N]=E[X]-E[R]=260\), which equals \(E[U]-E[Q]=340-80=260\). The common mistake of treating the expected recovery as if it were always matched to a loss would miss the excess recovery in Scenario 4.
Why the Signed and Uncovered Measures Differ
The signed net loss \(N\) records the financial position after the payment, including negative values when the recovery exceeds the loss. Uncovered loss \(U\), by contrast, cannot be negative. Their relationship follows directly from considering the two possible orderings of \(X\) and \(R\).
Then \(U=X-R\) and \(Q=0\), so \(N=X-R=U-Q\).
Then \(U=0\) and \(Q=R-X\), so \(N=X-R=-Q=U-Q\).
Since \(N=U-Q\) in either case, \(E[N]=E[U]-E[Q]=E[X]-E[R]\), provided the expectations exist.
This identity explains an important distinction. The difference \(E[X]-E[R]\) is the expected signed net loss, not the expected uncovered loss. If a contract can pay more than actual loss in some scenarios, \(E[U]\) exceeds \(E[X]-E[R]\) by \(E[Q]\). For an indemnity arrangement whose payment cannot exceed the covered portion of actual loss, \(Q=0\) under that contract definition, so these two expected measures coincide. Do not assume that restriction for an index-based payment.
Worked Example: Separating Expected Recovery from Protection Against Loss
A hypothetical portfolio has four equally likely joint scenarios. Losses and recoveries are in thousands of dollars. The index cover pays \(R=\min(500,\max(0,20(Z-50)))\). Determine expected uncovered loss, expected excess recovery, and the probability of an uncovered loss greater than 500.
| Probability | Actual loss \(X\) | Index \(Z\) |
|---|---|---|
| 0.25 | 100 | 40 |
| 0.25 | 300 | 70 |
| 0.25 | 700 | 80 |
| 0.25 | 900 | 60 |
Method. Evaluate the payout formula in each scenario, including its threshold and cap. The payment is zero for \(Z\le 50\); above 50, it is 20 thousand dollars per point, subject to a 500-thousand-dollar maximum.
For \(Z=40\), \(R=0\), so \(U=100\) and \(Q=0\). For \(Z=70\), \(R=20(20)=400\), so \(U=0\) and \(Q=100\). For \(Z=80\), the uncapped payment is 600, so \(R=500\), \(U=200\), and \(Q=0\). For \(Z=60\), \(R=200\), \(U=700\), and \(Q=0\). Each amount is in thousands.
The expected uncovered loss is \(0.25(100+0+200+700)=250\), or $250,000. Expected excess recovery is \(0.25(0+100+0+0)=25\), or $25,000. Uncovered loss exceeds 500 only in the final scenario, so \(P(U>500)=0.25\).
As a check, expected actual loss is \(0.25(100+300+700+900)=500\), and expected recovery is \(0.25(0+400+500+200)=275\), both in thousands. Thus expected signed net loss is \(500-275=225\). It is also \(E[U]-E[Q]=250-25=225\). A common incorrect calculation would report $225,000 as expected uncovered loss. That is the expected signed net loss; the correct expected uncovered loss is $250,000 because one scenario has an excess recovery.
Basis Risk Can Change the Shape of Retained Risk
A cover’s average payment does not establish whether it reduces variability in the insurer’s retained position. If payments are poorly aligned with losses, the payment can arrive when actual losses are small and fail to arrive when losses are large. The following hypothetical example uses the same scenario probabilities to compare the variance of actual loss with the variance of signed net loss.
Worked Example: When an Index Payment Increases Variability
Four joint scenarios are equally likely. Amounts are in thousands of dollars. The cover pays \(R=\min(400,\max(0,20(Z-60)))\). Calculate the scenario payments and signed net losses, then compare the variance of \(X\) with the variance of \(N=X-R\).
| Probability | Actual loss \(X\) | Index \(Z\) |
|---|---|---|
| 0.25 | 0 | 80 |
| 0.25 | 200 | 20 |
| 0.25 | 600 | 100 |
| 0.25 | 1,000 | 40 |
Method. First calculate \(R\), then \(N=X-R\) in each scenario. The payout is capped at 400 and is zero when the index is at most 60.
At \(Z=80\), the uncapped payment is \(20(20)=400\), so \(R=400\) and \(N=0-400=-400\). At \(Z=20\), \(R=0\) and \(N=200\). At \(Z=100\), the uncapped payment is 800, so \(R=400\) and \(N=600-400=200\). At \(Z=40\), \(R=0\) and \(N=1{,}000\). Thus the signed net losses are \(-400,200,200,1{,}000\).
The mean actual loss is \(E[X]=(0+200+600+1{,}000)/4=450\). The mean net loss is \(E[N]=(-400+200+200+1{,}000)/4=250\). For equally likely scenarios, variance is the average squared deviation from the relevant mean. Therefore:
The variance increases by \(247{,}500-147{,}500=100{,}000\) in squared thousands of dollars. The payment has not reduced variability in these scenarios: it pays $400,000 when the actual loss is zero, while the largest loss receives no payment. The corresponding expected uncovered loss is \((0+200+200+1{,}000)/4=350\), and expected excess recovery is \(400/4=100\), both in thousands. As a check, \(E[X]-E[R]=450-(400+0+400+0)/4=250\), equal to \(E[N]\) and to \(350-100\). This variance comparison is specific to the stated joint scenarios; it is not a general claim that index covers increase risk.
Common Traps in Measurement
- Using only expected values. A positive expected recovery does not reveal whether the payment arrives in high-loss scenarios. Compare \(X\) and \(R\) scenario by scenario.
- Using correlation as a complete basis-risk measure. Correlation is a summary of linear association. It does not directly report the probability of a shortfall, the size of shortfalls after a trigger, or the frequency of excess recoveries.
- Applying an indemnity constraint to an index payment. Do not replace \(R\) by \(\min(R,X)\) unless the contract actually limits payment to covered actual loss. Doing so erases excess recovery and changes the contract being analyzed.
- Confusing expected signed loss with expected uncovered loss. \(E[X]-E[R]\) includes negative net losses; \(E[(X-R)^+]\) does not. They differ by \(E[(R-X)^+]\).
- Ignoring units, probabilities, or caps. Convert all amounts to a common unit, verify that scenario probabilities sum to one, and apply the payout cap before calculating mismatch.
A useful assessment therefore reports more than a single average. At minimum, distinguish the expected uncovered loss from expected excess recovery, and examine the chance and size of uncovered losses in severe scenarios. The appropriate summaries depend on the purpose of the analysis, but they must reflect the joint outcomes of actual loss and the external trigger. This approach to identifying and quantifying basis risk is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
Check Your Understanding
For each problem, apply the stated payout rule to the given joint scenarios. Amounts are in thousands of dollars unless otherwise stated.
- A cover pays \(R=\min(300,\max(0,10(Z-20)))\). There are two equally likely scenarios: \((X,Z)=(80,10)\) and \((500,50)\). Find expected uncovered loss and expected excess recovery. (Answer: Expected uncovered loss = 140; expected excess recovery = 0.)
- A contract pays \(R=\min(300,\max(0,10(Z-20)))\). The scenarios are \((X,Z)=(80,10)\) with probability 0.50 and \((X,Z)=(500,50)\) with probability 0.50. Find expected uncovered loss and expected recovery. (Answer: Expected uncovered loss = 140; expected recovery = 150.)
- Three scenarios have probabilities 0.20, 0.50, and 0.30, with \((X,R)=(100,0),(400,250),(800,900)\). Calculate expected uncovered loss, expected excess recovery, and expected signed net loss. (Answer: Expected uncovered loss = 95; expected excess recovery = 30; expected signed net loss = 65.)
- A cover produces the following \((X,R)\) pairs: \((100,200)\) with probability 0.10, \((500,200)\) with probability 0.40, and \((900,500)\) with probability 0.50. Find expected uncovered loss and the probability that uncovered loss exceeds 250. (Answer: Expected uncovered loss = 320; probability = 0.90.)
- In a portfolio of equally likely scenarios, actual losses are \(0,200,600,1{,}000\), and signed net losses after an index payment are \(-400,200,200,1{,}000\). Calculate the means and variances of actual and signed net loss. (Answer: \(E[X]=450\), \(E[N]=250\), \(\operatorname{Var}(X)=147{,}500\), and \(\operatorname{Var}(N)=247{,}500\), with variances in squared thousands of dollars.)