Introduction
A risk can be addressed by deciding not to undertake the activity that creates it, or by changing how the activity is performed so losses become less likely or less severe. These are different decisions. Withdrawing a product may prevent new exposure from arising, while redesigning a process typically leaves the activity in place and reduces its risk. In either case, a sound assessment asks what exposure is removed, what remains, and whether the change creates other risks.
The distinction matters in enterprise risk management because a lower expected loss does not necessarily mean an exposure has been eliminated. Existing policies, contracts, transactions, or assets may continue to generate losses after an organization stops creating new exposure. The analysis below focuses on that distinction and on simple expected-loss calculations used to describe it.
What Is Changed by Each Decision?
Avoidance changes the boundary of the activity: the organization no longer takes on a specified exposure, or it exits that exposure. Reduction changes the conditions under which the exposure is managed. Examples include redesigning a process to prevent errors, adding review before a payment is released, or imposing transaction limits that constrain the size of a loss.
The categories describe the effect of the decision, not merely the name of a control. A safeguard might reduce the chance of a loss, while a decision to stop offering the affected transaction type might avoid that particular future exposure. A withdrawal can also be partial: the organization may stop new sales but still face obligations from policies already issued. Define the exposure, the time period, and which existing obligations remain before assigning a label.
| Decision | What changes? | Question to ask |
|---|---|---|
| Avoidance | The organization does not undertake a specified future activity or exposure. | Which future losses can no longer arise, and what existing exposure remains? |
| Likelihood reduction | The activity continues, but a loss event is less likely. | What is the event frequency or probability before and after the change? |
| Impact reduction | The activity continues, but a loss, if it occurs, is less severe. | How does the conditional loss amount change? |
| Combined reduction | Both the likelihood and the severity of loss decrease. | Have both changes been reflected in the calculation? |
Expected loss is a useful summary for the calculations that follow. For \(n\) comparable exposures, each with event probability \(p\) over the specified period and conditional mean loss \(m\) if an event occurs, expected aggregate loss is \(npm\). This calculation requires the stated probabilities and conditional means to describe the exposures being analyzed. It does not require independence to calculate the expectation, but it does not by itself describe the variability or tail of aggregate loss.
To justify the expression, let \(I_j\) equal 1 if exposure \(j\) produces an event and 0 otherwise, and let \(Z_j\) be its loss amount when an event occurs. Write the loss for exposure \(j\) as \(I_jZ_j\). If \(P(I_j=1)=p\) and the conditional mean of \(Z_j\), given an event, is \(m\), then \(E[I_jZ_j]=pm\). The aggregate loss is the sum over the \(n\) exposures. By linearity of expectation, its mean is the sum of the individual means, \(npm\). This explains why the formula can be used without assuming that the exposures are independent; independence would matter for other aggregate-loss quantities, such as variance.
Worked Examples
Worked Example: Withdrawing a Product With Existing Policies
A hypothetical insurer is considering withdrawing a product from new sales. If it continues selling, it expects 1,200 new policies during the coming year. For each new policy, the probability of a claim during that year is 0.002, and the mean claim amount conditional on a claim is $25,000. The insurer also has 400 existing policies still in force; for each, the probability of a claim during the coming year is 0.0015 and the conditional mean claim is $20,000. Calculate the expected claim amount in the coming year if sales continue and if the product is withdrawn from new sales, assuming the existing policies remain in force. Identify the expected loss avoided by stopping new sales.
Method. Separate the new-sales exposure from the existing-policy exposure, then apply \(npm\) to each group. The calculation assumes the stated probabilities and conditional mean claims remain applicable over the coming year.
For the new policies, the expected number of claims is \(1{,}200(0.002)=2.4\). The expected claim amount from this group is \(2.4($25{,}000)=$60{,}000\).
For the existing policies, the expected number of claims is \(400(0.0015)=0.6\). The expected claim amount from this group is \(0.6($20{,}000)=$12{,}000\).
If sales continue, expected claims from both groups total \($60{,}000+$12{,}000=$72{,}000\). If new sales stop and the existing policies remain in force, the new-sales exposure is removed but the existing-policy exposure remains, so expected claims are $12,000. The expected loss avoided by stopping new sales is therefore $60,000 for this coming year.
Check. The difference between the continue-sales and withdrawn-sales totals is \($72{,}000-$12{,}000=$60{,}000\), matching the expected claims from the new policies. A common mistake is to count withdrawal as eliminating all $72,000 of expected claims. That incorrectly treats the existing policies as if they disappeared; their $12,000 expected claims remain within the stated scope.
Worked Example: Redesigning a Claims Process
A hypothetical claims operation expects 18 processing errors per year under its current design. Each error has a mean loss of $8,000. A proposed process redesign is expected to reduce the number of errors by 40%, while leaving the mean loss per remaining error unchanged. Calculate the expected annual loss before and after redesign and the expected annual reduction. Do not include the cost of implementing the redesign.
Method. Model the given annual error counts as expected frequencies and multiply each by the conditional mean loss per error. This calculation isolates the risk effect of the redesign; it does not establish whether the redesign is worthwhile after costs or other consequences are considered.
Before redesign, expected annual loss is \(18($8{,}000)=$144{,}000\).
A 40% reduction leaves \(100\%-40\%=60\%\) of the expected errors. The expected number after redesign is \(18(0.60)=10.8\) errors per year. The resulting expected annual loss is \(10.8($8{,}000)=$86{,}400\).
The expected annual reduction is \($144{,}000-$86{,}400=$57{,}600\). Equivalently, the reduced frequency is \(18(0.40)=7.2\) fewer expected errors, and \(7.2($8{,}000)=$57{,}600\).
Check. Since severity is unchanged, a 40% decrease in expected frequency must produce a 40% decrease in expected loss. Indeed, \($57{,}600/$144{,}000=0.40\). A common mistake is to apply the 40% reduction to the error severity instead of the error frequency. The problem states that severity is unchanged; only the expected number of errors falls.
Worked Example: Tightening Transaction Safeguards
A hypothetical firm processes 1,000 transfers in a year. Before a safeguard change, each transfer has probability 0.004 of producing a fraudulent loss, with conditional mean loss of $50,000. A revised authorization process is expected to reduce that probability to 0.0015. A per-transfer release limit and a review step are also expected to reduce the conditional mean loss, given a fraudulent loss, to $20,000. Calculate expected annual loss before and after the change, and the expected reduction.
Method. Apply \(npm\) separately before and after the safeguards. The stated post-change conditional mean is already the loss amount given an event; do not substitute the original severity when calculating the post-change result.
Before the change, the expected number of fraudulent losses is \(1{,}000(0.004)=4\). Expected annual loss is \(4($50{,}000)=$200{,}000\).
After the change, the expected number of fraudulent losses is \(1{,}000(0.0015)=1.5\). Expected annual loss is \(1.5($20{,}000)=$30{,}000\).
The expected annual reduction is \($200{,}000-$30{,}000=$170{,}000\). This is an 85% reduction relative to the original expected loss, since \($170{,}000/$200{,}000=0.85\).
Check. The remaining loss is 15% of the original: the likelihood ratio is \(0.0015/0.004=0.375\), and the conditional mean ratio is \($20{,}000/$50{,}000=0.40\). Their product is \(0.375(0.40)=0.15\), so the new expected loss is \(0.15($200{,}000)=$30{,}000\). A common mistake is to use the old $50,000 conditional mean after the safeguards, obtaining \(1.5($50{,}000)=$75{,}000\). That calculation ignores the stated impact reduction.
Interpreting the Results
These calculations describe expected losses under specified assumptions, not guaranteed outcomes. An expected count such as 10.8 errors does not mean that a fraction of an error will occur; it is a mean across possible outcomes or repeated comparable periods. Actual counts and losses can be higher or lower. A change in expected loss also does not, on its own, establish the change in the probability of a severe aggregate loss.
For avoidance, be precise about the scope of what is stopped. Ending new sales can remove a future stream of exposure without cancelling in-force contracts. Exiting an activity may also create other consequences, such as transition or service obligations, that need separate identification. Do not silently assume that those exposures vanish merely because the organization has made an avoidance decision.
For reduction, identify whether the intervention changes likelihood, impact, or both. A control that prevents some events but leaves the size of each resulting loss unchanged reduces frequency, not conditional severity. A transaction limit may constrain severity, but it does not necessarily eliminate the chance of an event. Where a measure affects both, use the post-change assumptions for both factors, as in the transaction example.
Expected-loss calculations are useful for comparing the exposure described by the assumptions, but they do not capture every risk dimension. For example, a safeguard may introduce reliance on a new approval system, or an average loss may conceal a small chance of a much larger event. State what is included and avoid treating a lower expected loss as proof that all material risk has been removed. This analysis is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
Check Your Understanding
For each problem, distinguish what is removed from what remains, and use the stated probabilities and mean losses for the specified period.
- A hypothetical insurer stops accepting new contracts in a product line. The next year, 600 new contracts would otherwise each have a claim probability of 0.003 and a conditional mean claim of $10,000. The insurer also has 150 existing contracts, each with a claim probability of 0.002 and a conditional mean claim of $8,000. If existing contracts remain in force, what expected claim amount is avoided by stopping new sales, and what expected claims remain? (Answer: $18,000 avoided; $2,400 remain.)
- A hypothetical process has an expected 25 errors per year, each with a mean loss of $6,400. A redesign reduces expected error frequency by 40% and does not change severity. Find expected annual loss before and after redesign and the reduction. (Answer: $160,000 before, $96,000 after, and $64,000 reduction.)
- A firm processes 400 transfers annually. Each has a 0.005 probability of a loss, with conditional mean severity of $30,000. A safeguard reduces the probability to 0.002 without changing severity. Find expected annual loss before and after, and the reduction. (Answer: $60,000 before, $24,000 after, and $36,000 reduction.)
- For 2,000 hypothetical transactions, the initial event probability is 0.003 and the conditional mean loss is $15,000. After a safeguard, the event probability is 0.002 and the conditional mean loss is $9,000. Calculate expected loss in each case and the percentage reduction. (Answer: $90,000 before, $36,000 after, and a 60% reduction.)
- A company stops creating new exposure to a product, but remains obligated on contracts already issued. Is the decision avoidance or reduction with respect to new contracts, and what must the risk assessment still include? (Answer: Avoidance of the specified new-contract exposure; it must still include losses from existing contracts and any other obligations within scope.)