Avoiding Basis and Tail-Risk Errors in Pricing Cases

Why Pricing-Case Basis Errors Matter

A pricing calculation can use correct arithmetic and still be wrong because its quantities are on different bases. Historical losses may be at an earlier cost level than the target rates; exposure counts may refer to different time periods; and a risk load may be applied to a different amount than the pricing rule specifies. Confusing value at risk with tail value at risk can also understate the severity of losses beyond a percentile threshold.

The discipline is to label each amount before combining it: loss-cost date, exposure unit and period, target cost level, and risk-load basis. This tutorial uses hypothetical cases. The checks are aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

Definition: A basis error occurs when a calculation combines quantities that do not represent the same exposure unit, time period, cost level, or risk measure. A trend adjustment moves losses from a stated historical cost level to a stated target cost level. An exposure adjustment scales a loss cost per exposure unit to a different number of comparable exposure units.

A Structured Basis Audit

Before calculating a projected loss amount or premium, write down what each input represents. In particular, distinguish the date to which the historical losses are already adjusted from the date to which they must be projected. If the losses are already at the target cost level, applying trend again is a duplicate adjustment.

Identify the exposure unit and period.
For example, a policy-year is not interchangeable with a policy, and two years of exposure are not one year of exposure unless the calculation explicitly annualizes them.

Locate the historical loss-cost level.
Record whether losses are at the historical level or have already been trended to the target level.

Apply trend only for the remaining time interval.
Use the stated annual trend rate and number of years between the loss-cost date and target date.

Scale to the target exposure count.
First obtain a loss cost per comparable exposure unit; then multiply by the target number of units.

Apply the stated risk rule.
Check whether the load is defined on expected loss, capital, excess tail risk, or another explicit base.

Result: If historical losses $L_H$ are at a historical cost level, there are $E_H$ comparable historical exposure units, the annual compound trend rate is $t$ for $n$ years, and the target exposure count is $E_T$, then

L_T=\frac{L_H}{E_H}(1+t)^n E_T.

If $L_H$ is already at the target cost level, omit $(1+t)^n$. This formula assumes the historical and target exposure units are comparable and that a single trend factor appropriately represents the change in loss costs.

Worked Examples: Correcting the Basis Before Pricing

Worked Example: Trend Once, Then Adjust for Exposure

A hypothetical insurer observed $400,000 of losses from 8,000 earned policy-years. The losses are at the historical cost level. The target cost level is two years later, and annual compound trend is 4%. The target portfolio has 10,000 earned policy-years for the same coverage. Calculate projected target-level losses. Then determine what the projection would be if the $400,000 had already been adjusted to the target cost level.

Method. Separate the per-exposure loss cost, trend, and target exposure count. The given premise says the $400,000 is historical-level loss, so trend is applied once.

Step 1: Historical loss cost. Divide by the historical policy-years: $400{,}000/8{,}000=\$50$ per policy-year.

Step 2: Trend factor. Two years at 4% annual compound trend gives $(1.04)^2=1.0816$. The trended loss cost is $\$50(1.0816)=\$54.08$ per policy-year.

Step 3: Scale to target exposure. $\$54.08(10{,}000)=\$540{,}800$. Thus projected target-level losses are $540,800.

Check the alternative premise. If $400,000 were already at the target cost level, its target-level loss cost would be $\$400{,}000/8{,}000=\$50$ per policy-year. The projection would then be $\$50(10{,}000)=\$500{,}000$, with no additional trend factor. This is the key distinction: the same numerical input gives a different projection when its cost-level status differs.

Common mistake avoided. Applying 1.0816 to the $400,000 after stating that it is already at the target cost level would duplicate trend. Conversely, omitting trend when the losses are explicitly historical-level would leave the projection at the wrong cost level.

Worked Example: Apply a Tail Load to the Defined Basis

A hypothetical one-period aggregate loss $X$, measured in thousands of dollars, has this distribution: $X=0$ with probability 0.8000, $X=100$ with probability 0.1500, and $X=300$ with probability 0.0500. A pricing rule sets premium equal to expected loss plus 25% of the excess of 90% TVaR over expected loss. Find the premium.

Method. Compute expected loss and the specified TVaR first. Then apply the 25% factor to the excess of TVaR over the mean, not to TVaR itself.

Step 1: Expected loss. In thousands of dollars, $E[X]=0(0.8000)+100(0.1500)+300(0.0500)=15+15=30$. Expected loss is therefore $30,000.

Step 2: 90% VaR. The cumulative probability at 0 is 0.8000, and at 100 it is 0.9500. The smallest loss with cumulative probability at least 0.9000 is 100, so $\operatorname{VaR}_{0.90}(X)=\$100{,}000$.

Step 3: 90% TVaR. The worst 10% of outcomes contains all 5% of the probability at 300 and 5% of the probability at 100. Thus, in thousands, \[ \operatorname{TVaR}_{0.90}(X) =\frac{0.0500(300)+0.0500(100)}{0.1000} =\frac{15+5}{0.1}=200. \] So 90% TVaR is $200,000.

Step 4: Apply the stated load rule. The excess tail amount is $200-30=170$ thousand dollars. The risk load is $0.25(170)=42.5$ thousand dollars. Premium is $30+42.5=72.5$ thousand dollars, or $72,500.

Check. Expected loss plus the calculated load is $30{,}000+42{,}500=\$72{,}500$, as required. A common incorrect calculation is to use $0.25(200{,}000)=\$50{,}000$ as the load and report a $80,000 premium. That applies the factor to the full TVaR, not to the excess over expected loss specified in the rule.

VaR and TVaR Are Different Tail Summaries

Definition: For a loss $X$, the $q$-level value at risk is the quantile $\operatorname{VaR}_q(X)=\inf\{x:F_X(x)\ge q\}$. Tail value at risk is the average of the quantile function over the upper tail:

\operatorname{TVaR}_q(X)=\frac{1}{1-q}\int_q^1 \operatorname{VaR}_p(X)\,dp.

For a discrete distribution, this definition includes only the needed fraction of probability mass at the boundary quantile. It is not always the same as simply averaging all outcomes strictly greater than $\operatorname{VaR}_q(X)$.

VaR identifies a threshold: at least probability $q$ lies at or below it. TVaR summarizes the average loss in the worst $1-q$ fraction of outcomes, with fractional inclusion at a discrete boundary if necessary. Therefore, in general, TVaR is not the VaR threshold and cannot be substituted for it, or vice versa, without changing the stated pricing or capital rule.

Worked Example: Do Not Substitute VaR for TVaR

For a hypothetical portfolio, the one-period aggregate loss $X$, in thousands of dollars, is 0 with probability 0.9000, 100 with probability 0.0900, and 1,000 with probability 0.0100. A pricing rule is $E[X]+10\%\bigl(\operatorname{TVaR}_{0.95}(X)-E[X]\bigr)$. Calculate the premium and show the effect of mistakenly using VaR in place of TVaR.

Step 1: Expected loss. $E[X]=0(0.9000)+100(0.0900)+1{,}000(0.0100)=9+10=19$ thousand dollars.

Step 2: VaR at 95%. The cumulative probability at zero is 0.9000, while at 100 it is 0.9900. Hence $\operatorname{VaR}_{0.95}(X)=\$100{,}000$.

Step 3: TVaR at 95%. The worst 5% consists of the 1% at 1,000 and 4% of the probability at 100. In thousands, $\operatorname{TVaR}_{0.95}(X)=[0.0100(1{,}000)+0.0400(100)]/0.0500=(10+4)/0.05=280$. Therefore TVaR is $280,000.

Step 4: Apply the rule. The TVaR excess over expected loss is $280-19=261$ thousand dollars. The load is $0.10(261)=26.1$ thousand dollars, and premium is $19+26.1=45.1$ thousand dollars, or $45,100.

Wrong calculation and correction. Substituting VaR gives $19+0.10(100-19)=27.1$ thousand dollars, or $27,100. That is not the requested TVaR-based premium: it replaces the average of the worst 5% with the 95th-percentile threshold. The correct premium under the stated rule is $45,100.

Why the Formulas Work

The exposure and trend formula follows from the meaning of a loss cost. If $L_H$ is the total loss from $E_H$ comparable exposure units, the historical average loss per unit is $L_H/E_H$. Under the stated trend assumption, the target-level expected loss per unit is $(L_H/E_H)(1+t)^n$. Multiplying by $E_T$ target units gives the projected total. This scaling requires comparable units and a stable enough expected loss cost per unit; it does not justify scaling across materially different coverage or risk mixes without an adjustment.

The discrete TVaR calculation also follows directly from the quantile definition. At confidence level $q$, the upper tail has probability $1-q$. If the boundary loss value has more probability mass than is needed to complete that tail, only the required fraction of that mass belongs in the tail average. In the 90% example above, the 5% at 300 fills half of the worst 10%; the remaining half comes from the 100 outcome. Dividing the tail-weighted loss total by 0.10 produces its average.

Exam Traps and Practical Checks

A quick audit often catches errors before they spread through a premium or capital recommendation. Keep the following distinctions explicit:

Policy versus policy-year: A count of policies is not automatically a count of annual exposure units. If exposure spans different durations, use the exposure measure stated in the pricing basis.
Historical level versus target level: Trend only losses that have not already been adjusted to the target level. Identify the start and end cost dates before choosing the number of trend periods.
Trend versus exposure growth: Trend changes the loss cost per exposure unit. Increasing the target exposure count changes the total number of units. These are separate operations.
VaR versus TVaR: VaR is a quantile threshold; TVaR is an average over the upper tail. State which one the rule uses.
Load base: Read the rule algebraically. A percentage of $ \operatorname{TVaR}_q(X)-E[X]$ differs from the same percentage of $\operatorname{TVaR}_q(X)$.

These checks diagnose calculation consistency; they do not establish that the selected trend, exposure measure, confidence level, or load rule is appropriate for every portfolio. Those assumptions must come from the problem statement or the pricing basis being evaluated.

Key takeaway: Write each input with its unit, time period, and cost level; apply each adjustment once; and follow the stated tail-load formula exactly. A correct pricing result depends as much on matching the bases as on performing the arithmetic.

Check Your Understanding

For each hypothetical case, identify the correct basis and calculate the requested amount.

Historical-level losses are $275,000 for 5,000 exposure-years. Trend is 3% for one year to the target cost level, and the target has 6,000 exposure-years. What are projected target-level losses? (Answer: $339,900)
Losses of $720,000 for 12,000 policy-years are already at the target cost level. The target portfolio has 15,000 comparable policy-years. What projected losses result if no further trend is needed? (Answer: $900,000)
A loss is 0 with probability 0.7000, $80,000 with probability 0.2000, and $200,000 with probability 0.1000. Find $\operatorname{VaR}_{0.80}$ and $\operatorname{TVaR}_{0.80}$. (Answer: VaR = $80,000; TVaR = $140,000)
A pricing rule specifies expected loss plus 20% of the excess of TVaR over expected loss. If expected loss is $50,000 and TVaR is $170,000, find the risk load and premium. (Answer: Risk load = $24,000; premium = $74,000)
Historical-level losses are $1,080,000 from 18,000 exposure-years accumulated over a two-year period. Treat those as 18,000 total exposure-years, not 18,000 per year. The target is 12,000 exposure-years, two years after the stated loss-cost date, with annual compound trend of 5%. Find target-level losses. (Answer: $793,800)

What You'll Learn