Why Strategic Alternatives Need More Than One Measure
Strategic choices rarely differ on only one dimension. An initiative may promise greater expected benefit but have a wider range of outcomes, a more severe downside, or weaker alignment with an organization’s priorities. A comparison that reports only expected benefit can hide these differences; a single composite score can hide them in a different way.
A useful comparison makes the trade-offs visible. It puts alternatives on a consistent basis, describes what could go well and badly, and identifies which organizational priorities or constraints matter to the decision. This approach is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
Before calculating, define the outcome measure and comparison basis. For example, decide whether \(B_j\) is annual incremental operating contribution, total cost over a fixed period, or another measure. Alternatives should use the same horizon, currency, baseline, and treatment of costs. If one alternative’s benefits are measured before implementation costs and another’s after costs, the comparison is not meaningful.
A Structured Comparison
For an alternative with possible outcomes \(b_{jk}\), where scenario \(k\) has probability \(p_{jk}\), the expected benefit is the probability-weighted average. The probabilities for one alternative must be nonnegative and sum to one.
Expected benefit is a useful summary, but two alternatives with the same expected benefit may have very different downside exposure. Other descriptions can include the probability that benefit is below a stated threshold \(T\), and the expected shortfall below that threshold. Define shortfall as \(\max(T-B_j,0)\); it is zero whenever the outcome meets or exceeds \(T\).
These are distinct summaries. A downside probability says how often a threshold is missed, while expected shortfall also reflects how far below it the outcomes lie. The threshold must be chosen for a reason, such as a minimum acceptable contribution or an approved capacity constraint; changing it can change the comparison.
Uncertainty can also be described by a range or, where useful, the standard deviation of the outcome. For a discrete model, \(\operatorname{Var}(B_j)=\sum_k p_{jk}(b_{jk}-E[B_j])^2\), and the standard deviation is the square root of that variance. These measures summarize spread around the mean; they do not distinguish favorable variation from unfavorable variation. They should not replace explicit downside measures when downside is central to the decision.
Why Expected Benefit Is a Weighted Average
The expected-value formula follows from the definition of expectation for a discrete outcome. Imagine repeating the decision many times under the same scenario probabilities. In the long run, the proportion of occasions in scenario \(k\) approaches \(p_{jk}\). The average outcome is therefore the sum of each scenario’s outcome multiplied by its long-run proportion.
Alternative \(j\) produces \(b_{jk}\) in scenario \(k\), with probability \(p_{jk}\).
Across many repetitions, scenario \(k\) contributes approximately \(p_{jk}b_{jk}\) to the average per decision.
The average across all scenarios is \(\sum_k p_{jk}b_{jk}\), which is the expected benefit.
This derivation explains both the value and the limit of the measure: expected benefit summarizes the average under the specified probabilities, but it cannot say whether an individual outcome is likely to be poor, whether the probabilities are reliable, or whether the alternative meets a strategic constraint.
Worked Comparisons
Worked Example: Comparing Average Benefit and Downside
A hypothetical insurer is considering two service expansions. For each alternative, annual incremental contribution is measured in millions of dollars, after ordinary operating costs, relative to the same baseline. Assume the following outcomes and probabilities are complete and mutually exclusive. Compare expected contribution, probability of a negative outcome, expected shortfall below zero, and standard deviation.
Alternative A produces \(-2\) with probability \(0.20\), \(4\) with probability \(0.50\), and \(8\) with probability \(0.30\). Alternative B produces \(1\), \(3\), or \(5\) with probabilities \(0.20\), \(0.50\), and \(0.30\), respectively.
Method. Calculate each measure from the scenario distribution, using zero as the shortfall threshold. This avoids mistaking a larger average for a complete description of the choice.
For A, the expected contribution is \(0.20(-2)+0.50(4)+0.30(8)=-0.40+2.00+2.40=4.00\) million dollars. For B, it is \(0.20(1)+0.50(3)+0.30(5)=0.20+1.50+1.50=3.20\) million dollars.
A has a negative outcome only in its \(-2\) scenario, so \(P(B_A<0)=0.20\). Its expected shortfall below zero is \(0.20(2)+0.50(0)+0.30(0)=0.40\) million dollars. B has no negative outcome, so its negative-outcome probability and expected shortfall below zero are both zero.
For A, the deviations from the mean of \(4\) are \(-6,0,4\). Its variance is \(0.20(36)+0.50(0)+0.30(16)=7.20+4.80=12.00\) million dollars squared, giving standard deviation \(\sqrt{12}=3.46\) million dollars, rounded to two decimals. For B, the deviations from \(3.20\) are \(-2.20,-0.20,1.80\). Its variance is \(0.20(4.84)+0.50(0.04)+0.30(3.24)=0.968+0.020+0.972=1.960\), so its standard deviation is \(1.40\) million dollars.
Conclusion and check. A has the higher expected contribution, \(4.00\) versus \(3.20\) million dollars, but also greater spread and a \(20\%\) chance of a negative contribution. B has no negative scenario in the stated model. The calculations are consistent with the listed distributions: each set of probabilities sums to \(1.00\). Which is preferable depends on priorities and constraints, not on the expected value alone.
Common mistake: choosing A solely because \(4.00>3.20\). That ignores its negative outcome. The correction is to report average benefit alongside downside and then assess those measures against the organization’s priorities.
Worked Example: Applying a Hard Guardrail Before Weighing Benefits
A hypothetical mutual insurer compares two initiatives over the same one-year horizon. Net contribution, in millions of dollars after ordinary implementation and operating costs, is scenario-dependent. The organization has set a hard requirement that no modeled scenario may produce contribution below \(2\) million dollars. Subject to that requirement, it seeks expected contribution of at least \(4\) million dollars. Initiative A produces \(1,5,7\) with probabilities \(0.25,0.50,0.25\); initiative B produces \(2,4,6\) with probabilities \(0.10,0.60,0.30\). Determine which initiatives satisfy the requirements.
Method. Test the minimum-outcome guardrail first, then calculate expected contribution for any alternative that passes. A hard constraint is not simply another preference that can be offset by a high average.
For A, the minimum modeled outcome is \(1\) million dollars, which is below the \(2\) million dollar guardrail. A fails, regardless of its expected contribution. Its expected contribution is \(0.25(1)+0.50(5)+0.25(7)=0.25+2.50+1.75=4.50\) million dollars, so it would meet the average requirement but still fail the hard constraint.
For B, the minimum outcome is \(2\) million dollars, so it meets the guardrail exactly. Its expected contribution is \(0.10(2)+0.60(4)+0.30(6)=0.20+2.40+1.80=4.40\) million dollars. This exceeds the \(4\) million dollar average requirement by \(0.40\) million dollars.
Conclusion and check. B satisfies both requirements; A satisfies only the expected-contribution requirement. Each probability set sums to \(1.00\). Note that the guardrail is based on modeled scenarios: passing it does not establish that outcomes outside the model are impossible.
Common mistake: treating A’s \(4.50\) million dollar expected contribution as compensation for an outcome below the hard minimum. The correction is to apply genuine non-negotiable constraints as eligibility tests before comparing preferences among eligible alternatives.
Worked Example: Making Organizational Priorities Visible
A hypothetical insurer is choosing between two claims-service platforms. Each has a one-year expected net contribution, in millions of dollars, after ordinary costs. Platform A has expected contribution \(6.20\); platform B has expected contribution \(5.20\). A scenario analysis also gives the following downside information, measured against zero contribution: A has a \(0.10\) probability of a \(4\) million dollar loss, with no other negative outcome; B has a \(0.10\) probability of a \(1\) million dollar loss, with no other negative outcome. Separately, a documented review rates A’s fit with the organization’s priority of reliable service access as moderate, and B’s fit as strong. The organization’s approved maximum modeled loss is \(3\) million dollars. Recommend how to compare them.
Method. Keep the financial measures, the hard loss limit, and the strategic fit distinct. Do not convert the qualitative ratings into arbitrary numbers and add them to contribution.
For A, the probability of a negative outcome is \(0.10\), and the expected shortfall below zero is \(0.10(4)=0.40\) million dollars. For B, the probability is also \(0.10\), but expected shortfall is \(0.10(1)=0.10\) million dollars. A’s modeled loss of \(4\) million dollars exceeds the \(3\) million dollar limit, so A fails the approved constraint. B’s modeled loss of \(1\) million dollars is within the limit.
The contribution comparison favors A by \(6.20-5.20=1.00\) million dollars in expected annual contribution. However, A does not meet the loss limit and has only moderate fit with the stated priority, while B meets the limit and has strong fit. On the information given, B is the eligible choice. If the organization revisits A, it would need to establish a feasible way to bring its modeled exposure within the limit and reassess the evidence on service access; the higher expected contribution alone does not resolve those issues.
Conclusion and check. The recommendation follows from the stated hard limit and priority, not from claiming that B is superior on every measure. The expected-contribution difference is \(1.00\) million dollars, and the expected shortfalls are correctly calculated as \(0.40\) and \(0.10\) million dollars.
Common mistake: inventing a numerical score for “moderate” and “strong” fit, then combining it with dollars. The correction is to report the ratings with their evidence and decision relevance, while keeping quantitative and qualitative criteria interpretable.
Using the Comparison in a Decision
A clear comparison distinguishes constraints from preferences. A constraint is a condition an alternative must satisfy, such as a maximum modeled loss or a minimum service standard. A preference helps rank alternatives that remain feasible, such as stronger alignment with an approved strategic objective. If a condition is truly mandatory, do not let a favorable result on another measure silently cancel it.
For preferences, a compact scorecard can help decision-makers see where alternatives differ. It should state the criteria, evidence, and direction of preference. Qualitative judgments should be supported by reasons—for example, which customer groups are affected or what capability is developed—and should not imply precision they do not possess. If the organization uses weights, show them and explain their basis; also check whether modest changes in weights would reverse the recommendation.
| Comparison dimension | Useful question | Interpretation |
|---|---|---|
| Expected benefit | What is the probability-weighted outcome on a common basis? | An average under stated assumptions, not a guaranteed result |
| Uncertainty | How widely do modeled outcomes vary? | Describes spread; does not by itself indicate whether outcomes are favorable or adverse |
| Downside | How likely and how severe are outcomes below a relevant threshold? | Requires an explicit threshold and attention to both probability and magnitude |
| Strategic fit | Which approved objectives does the alternative advance or hinder? | Needs contextual evidence, not an unexplained numerical label |
| Feasibility and constraints | Can the organization implement it, and does it meet mandatory limits? | May determine eligibility before preferences are weighed |
Scenario probabilities are estimates, not facts. When results depend materially on a probability, cost, or implementation assumption, show how the conclusion changes if that assumption changes. Also distinguish variability under the model from uncertainty about whether the model and its probabilities are appropriate. A precise calculation cannot correct incomplete scenarios or unsupported assumptions.
Common Comparison Traps
- Using inconsistent bases. Compare the same horizon, baseline, currency, and cost treatment; otherwise an apparent advantage may come from the measurement setup.
- Reporting only the average. An average does not show the chance or severity of a poor outcome. Add downside measures tied to a stated threshold.
- Treating spread as downside. Variance reflects deviations in both directions. It does not say which alternative has the more serious adverse outcomes.
- Blending a hard limit into a score. A mandatory constraint should be tested directly; a high value elsewhere does not make a breach disappear.
- Assigning false precision to strategic fit. Use a transparent rationale for qualitative judgments rather than presenting unsupported scores as objective measurements.
Check Your Understanding
For each hypothetical decision, use the stated horizon, outcome basis, and probabilities. Keep the calculation and the decision criterion distinct.
- A regional insurer compares two service investments, with annual net contribution in millions of dollars. A produces \(0,4,8\) with probabilities \(0.20,0.50,0.30\); B produces \(2,3,5\) with the same probabilities. Calculate each expected contribution and probability of contribution below \(2\). Which has the higher expected contribution, and which meets a requirement that every modeled outcome be at least \(2\)? (Answer: A has expected contribution \(4.40\) million dollars and a \(0.20\) probability below \(2\); B has expected contribution \(3.40\) million dollars and probability \(0\) below \(2\). A has the higher expectation, but only B meets the every-scenario minimum.)
- A hypothetical initiative has outcomes \(-3,2,6\) million dollars with probabilities \(0.10,0.40,0.50\). Calculate expected benefit and expected shortfall below zero. (Answer: Expected benefit is \(-0.30+0.80+3.00=3.50\) million dollars; expected shortfall below zero is \(0.10(3)=0.30\) million dollars.)
- Two alternatives have the same stated horizon and net-benefit basis. C produces \(1\) or \(7\) million dollars with probabilities \(0.50,0.50\); D produces \(3\) or \(5\) million dollars with probabilities \(0.50,0.50\). Calculate their expected benefits and standard deviations. If a decision-maker cares about spread but has not stated whether favorable variation is valued differently from adverse variation, what can be concluded? (Answer: Both expected benefits are \(4\) million dollars; C has standard deviation \(3\) million dollars and D has standard deviation \(1\) million dollar. D has less spread, but standard deviation alone does not establish which is preferable.)
- A proposed system has a \(0.15\) probability of a \(2\) million dollar loss and a \(0.85\) probability of a \(6\) million dollar gain. The organization’s maximum modeled loss is \(1.5\) million dollars. Calculate expected benefit and decide whether the system meets the loss constraint. (Answer: Expected benefit is \(0.15(-2)+0.85(6)=-0.30+5.10=4.80\) million dollars. It does not meet the constraint because the modeled loss of \(2\) million dollars exceeds \(1.5\) million dollars.)
- A board compares two hypothetical expansion plans. Plan M has higher expected contribution but breaches an approved maximum-loss limit; Plan N meets the limit and more directly supports an approved customer-access objective. What information should a decision summary keep separate, and what follows if the limit is mandatory? (Answer: Show expected contribution, downside against the limit, and evidence of customer-access alignment as distinct items. If the limit is mandatory, M is not eligible as modeled; N is eligible on the facts given, though other relevant criteria still need review.)