Managerial Decisions under Risk and Uncertainty

Introduction

Managers rarely possess complete information when making decisions. Human resource managers must forecast employee turnover, assess uncertain training benefits, choose workforce structures and evaluate technologies whose future effects are unknown. Business economics provides systematic methods for structuring these choices, measuring exposure and selecting actions consistent with organizational objectives and risk tolerance.

A decision is made under certainty when each action has a known outcome. Under risk, possible outcomes and defensible probabilities are available. Under uncertainty, outcomes, probabilities or both are insufficiently known. This distinction matters because probability-based methods are appropriate only when their assumptions are credible.

Risk analysis does not eliminate uncertainty. It makes assumptions visible, compares alternatives consistently and supports accountable managerial judgment.

Learning Objectives

After studying this chapter, students should be able to:

1. distinguish certainty, risk, uncertainty and ambiguity;
2. construct payoff tables and decision trees;
3. calculate expected monetary value, variance, standard deviation and coefficient of variation;
4. apply expected utility and risk-adjusted decision rules;
5. use sensitivity analysis, scenario analysis and Bayesian updating;
6. calculate the expected value of information;
7. analyze HR decisions involving uncertain costs and benefits; and
8. evaluate behavioral, ethical and governance limitations of quantitative models.

1. The Structure of a Managerial Decision

A managerial decision contains four basic elements:

• Decision alternatives: actions controlled by management, such as hiring permanent employees or contractors.
• States of nature: external conditions that management cannot fully control, such as strong, moderate or weak demand.
• Payoffs: economic consequences associated with each action–state combination.
• Probabilities: numerical assessments of how likely the states are, where credible estimates are available.

A compact representation is a payoff table. If a_i denotes action i, s_j denotes state j and x_ij is the monetary payoff, each table cell reports the outcome from choosing a_i when s_j occurs.

Payoffs must be defined consistently. Analysts should not compare one alternative using revenue and another using profit. Relevant payoffs may include recruitment costs, wages, severance obligations, productivity, turnover costs, employee well-being and reputational effects. Monetary models should be supplemented when important outcomes cannot responsibly be converted into money.

2. Probability and Sources of Risk

2.1 Objective and subjective probabilities

An objective probability is estimated from repeatable observations, such as historical absenteeism rates. A subjective probability is a reasoned judgment based on expertise, surveys or market intelligence. Most managerial forecasts combine both.

For mutually exclusive and collectively exhaustive states:

Here, p_j is the probability of state j, and n is the number of possible states.

Historical frequencies are informative only when the underlying environment is sufficiently stable. Structural changes—including automation, labor regulation, demographic shifts and economic crises—can make past data unrepresentative.

2.2 Risk versus deep uncertainty

Under measurable risk, managers can assign useful probabilities. Under deep uncertainty, decision makers may disagree about models, probability distributions or even the relevant outcomes. Robust strategies, stress tests and adaptive plans are often more defensible than precise expected-value calculations in such situations.

3. Expected Monetary Value

The expected monetary value (EMV) of action a_i is the probability-weighted average of its possible monetary payoffs:

Here, X_i is the uncertain payoff from action i, p_j is dimensionless and x_ij is measured in currency. EMV therefore has the same currency unit as the payoffs.

The EMV rule selects the action with the highest expected payoff, assuming outcomes are monetary, probabilities are credible and the organization is effectively risk-neutral over the relevant range.

Worked Example: Recruitment Campaigns

A company is comparing small and large recruitment campaigns. Four labor-market scenarios are judged equally likely.

The expected payoffs are:

A risk-neutral manager prefers the large campaign because its expected payoff is higher. This comparison, however, does not reveal the dispersion of outcomes.

4. Measuring Risk

4.1 Variance and standard deviation

For a discrete payoff distribution, population variance is:

Standard deviation is:

Variance is measured in currency squared, while standard deviation has the same currency unit as the payoff. These formulas describe the stipulated probability distribution; they are not the sample-variance formula used to estimate variance from a sample.

For the small campaign, deviations from $60,500 are $28,500, −$29,500, $19,500 and −$18,500:

For the large campaign, deviations from $100,000 are $98,000, −$98,000, $48,000 and −$48,000:

The large campaign offers a higher expected payoff but substantially greater absolute risk.

4.2 Coefficient of variation

When expected payoffs differ in scale, relative risk may be measured by the coefficient of variation (CV):

The CV is dimensionless. It is usually meaningful when μ_i > 0 and alternatives use comparable payoff definitions. It becomes unstable near zero and may be misleading when means are negative.

For the campaigns:

Thus, the large campaign has more risk per dollar of expected payoff.

4.3 A risk-adjusted value rule

Some organizations use a transparent risk penalty:

Here, k ≥ 0 is a dimensionless managerial risk-tolerance parameter. This is a policy rule, not a universal theorem. It should not be interpreted as a certainty equivalent unless justified by an appropriate utility model.

If k = 0.5:

Under this penalty, the large campaign remains preferred. Another value of k could change the ranking, so the parameter must reflect approved organizational risk appetite rather than analyst convenience.

5. Risk Attitudes and Expected Utility

Expected monetary value assumes risk neutrality. Managers and stakeholders may instead be:

• risk-averse: prefer a certain payoff to a risky payoff with the same mean;
• risk-neutral: evaluate alternatives by expected monetary payoff; or
• risk-seeking: prefer the risky alternative when expected payoffs are equal.

Expected utility theory represents preferences using a utility function U(x). The preferred action maximizes:

For a risk-averse decision maker, U′(x) > 0 and U″(x) < 0: utility increases with wealth but at a decreasing rate.

The certainty equivalent (CE) is the guaranteed payoff producing the same utility as the risky prospect:

The risk premium is:

For a risk-averse decision maker, the risk premium is normally positive.

A frequently used special result applies when utility is exponential and payoff is normally distributed. With constant absolute risk aversion coefficient A measured per currency unit:

This formula is not generally valid for arbitrary utility functions or non-normal payoff distributions. Mean–variance analysis may also overlook skewness, severe downside losses and tail risk.

6. Decisions When Probabilities Are Unavailable

When probabilities cannot be defended, several criteria can structure—not resolve—the decision.

For the Hurwicz criterion:

Here, 0 ≤ α ≤ 1 is the coefficient of optimism. The rule uses only the best and worst payoffs and therefore ignores intermediate outcomes.

Regret for action i in state j is:

Minimax regret selects the action with the smallest maximum regret. None of these rules is universally superior; each embodies a different attitude toward missing probability information.

7. Decision Trees and Sequential Choices

Decision trees are useful when decisions and uncertain events occur in sequence. Squares conventionally represent decision nodes, circles represent chance nodes and terminal branches show payoffs. Analysts evaluate a tree by rollback: calculate expected values at chance nodes and select the best branch at decision nodes.

Figure: A workforce decision tree separating managerial choices from uncertain demand states.

Probabilities along branches emerging from the same chance node must sum to one. Costs incurred before a chance event should be included once, not duplicated across stages.

8. Sensitivity and Scenario Analysis

8.1 Sensitivity analysis

Sensitivity analysis asks how the preferred action changes when one or more inputs vary. Suppose an HR analytics platform requires a fixed implementation cost of $60,000. Its gross incremental benefits are $150,000 under rapid adoption, $75,000 under moderate adoption and zero under failed adoption. Let p and q be the probabilities of rapid and moderate adoption, respectively, with p ≥ 0, q ≥ 0 and p + q ≤ 1.

The platform has positive expected net value when:

Therefore, the correct break-even condition is:

For example, if q = 0.40, adoption is worthwhile when:

This threshold shows precisely which probability judgments are decision-critical.

8.2 Scenario analysis and stress testing

Scenario analysis changes several related assumptions together. A recession scenario might combine low sales, hiring freezes, higher employee anxiety and lower voluntary turnover. Stress testing examines severe but plausible conditions, such as a sudden skills shortage or regulatory restriction.

Scenarios should be internally coherent rather than arbitrary combinations of best and worst values. Managers should also identify early warning indicators and pre-agreed responses.

9. Information, Bayesian Updating and Learning

9.1 Bayesian updating

New evidence should revise prior probabilities systematically. If H is a hypothesis and E is observed evidence, Bayes’ theorem is:

For mutually exclusive and exhaustive hypotheses H₁,…,H_n:

The denominator must be positive. A prior probability should represent knowledge before observing the current evidence; using the same evidence both to create the prior and update it causes double counting.

9.2 Expected value of perfect information

Perfect information reveals the state before a decision is selected. Its expected value is:

where:

EVPI cannot be negative because management can ignore information. It is the maximum rational price for perfect information under the stated monetary and risk-neutral assumptions.

The expected value of sample information (EVSI) measures the value of an imperfect survey, pilot or forecast. Its net value is positive only if EVSI exceeds the cost of obtaining and using the information. Imperfect information cannot be worth more than perfect information under consistent assumptions:

10. Illustrative Case: Northstar Workforce Expansion

Background

Northstar Services, a fictional business-process company, expects a new client contract but faces uncertain workload. HR must recommend either a permanent specialist team or contingent workers. Management evaluates annual net benefits after compensation, recruitment, training and coordination costs.

Problem

The permanent team promises greater average productivity but creates fixed employment obligations. Four demand-productivity scenarios are considered equally likely.

Analysis

For the permanent team:

Its deviations from the mean are $162,500, $62,500, −$107,500 and −$117,500. Therefore:

For contingent workers:

The deviations are $60,000, −$60,000, $20,000 and −$20,000:

EMV favors the permanent team by $15,000, but contingent workers have much lower variability. Under the illustrative rule RAV = μ − 0.5σ:

Approach and Recommendation

Northstar should not choose solely from a single metric. Management should:

1. validate scenario probabilities and payoff estimates;
2. quantify severance, knowledge-retention and contractor coordination costs;
3. test results under alternative demand probabilities;
4. consider a pilot or staged hiring strategy; and
5. assess employee fairness, legal classification and service-quality risks.

If the organization is risk-neutral and can absorb adverse outcomes, the permanent team is preferred by EMV. If downside protection and flexibility are strategic priorities, contingent staffing is more attractive. A hybrid approach—initial contingent capacity followed by permanent hiring after demand is confirmed—creates managerial flexibility.

Outcome and Learning Points

The case demonstrates that the highest expected payoff need not be the preferred risk-adjusted choice. Variance calculations must be interpreted alongside financial capacity, strategic flexibility and employment responsibilities. The analysis also shows why assumptions about risk tolerance should be explicit.

11. Real Options and Flexible Decisions

A real option is the managerial right, but not the obligation, to take a future action. Workforce examples include delaying recruitment, expanding a pilot, redeploying employees or abandoning a project.

Traditional EMV analysis may undervalue flexibility when decisions can be revised after learning. A staged hiring program can sacrifice some immediate efficiency while preserving the option to expand. Real-option analysis is especially useful when uncertainty is high, learning occurs over time and decisions are partly reversible.

Flexibility is not free. Delaying recruitment may cause lost output, employee overload or scarce-skill shortages. The value of an option must therefore be compared with its acquisition and exercise costs.

12. Behavioral Influences on Risk Decisions

Actual decision makers do not always follow expected utility theory. Common biases include:

• overconfidence: excessively narrow forecasts;
• anchoring: undue reliance on an initial number;
• availability bias: overweighting vivid or recent events;
• loss aversion: losses have greater psychological impact than equivalent gains;
• framing effects: choices change when identical outcomes are described differently;
• escalation of commitment: continuing a failing initiative because resources have already been spent;
• groupthink: suppressing disagreement to preserve consensus.

Debiasing practices include independent forecasts, premortem analysis, reference-class data, red teams, documented assumptions and separating project advocacy from project evaluation.

13. Ethical, Legal and Governance Considerations

Risk models in HR affect people, not merely cash flows. A staffing strategy with favorable expected cost may shift instability onto workers. Managers should consider distributive fairness, job security, workload, occupational health, privacy and access to development opportunities.

Algorithmic forecasts may reproduce historical discrimination. Governance should include:

• lawful and necessary data collection;
• data-quality and representativeness checks;
• testing for disparate impact;
• human review of consequential decisions;
• model documentation and version control;
• appeal and correction mechanisms; and
• periodic monitoring for model drift.

Probability estimates should not be manipulated to justify a predetermined decision. Analysts must disclose uncertainty ranges, excluded consequences and conflicts of interest. Compliance with employment, equality, privacy and worker-classification laws is a minimum requirement rather than a complete ethical standard.

14. Technology, AI and Contemporary Practice

Predictive analytics and artificial intelligence can estimate turnover, absenteeism, staffing demand and candidate supply. Monte Carlo simulation can generate many outcomes by repeatedly sampling uncertain inputs, producing an estimated payoff distribution rather than a single forecast.

These tools improve scale but do not remove judgment. Their reliability depends on data relevance, model assumptions and changing organizational conditions. Generative AI may assist scenario development, but unverified scenarios can be fictitious, biased or internally inconsistent. Human experts must validate causal logic, legal constraints and numerical outputs.

Contemporary practice increasingly emphasizes decision intelligence: integrating economic models, data analytics, experimentation, behavioral insight and governance. Organizations also use rolling forecasts and adaptive triggers because annual static plans respond poorly to rapidly changing labor markets.

15. Managerial Best Practices

A defensible risk decision process should:

1. define the decision, time horizon and decision owner;
2. identify feasible alternatives and mutually exclusive states;
3. use consistent payoff definitions and units;
4. document probability sources and confidence limits;
5. calculate expected values and risk measures accurately;
6. examine downside outcomes and distribution shape;
7. conduct sensitivity, scenario and break-even analysis;
8. value additional information before purchasing it;
9. identify reversible, staged or flexible alternatives;
10. review ethical, legal and workforce consequences;
11. record the rationale and approval process; and
12. monitor outcomes and update assumptions as evidence arrives.

Chapter Summary

Managerial decision analysis distinguishes certainty, measurable risk and uncertainty. Payoff tables and decision trees organize alternatives, external states, probabilities and consequences. Expected monetary value identifies the highest average monetary payoff under risk neutrality, while variance, standard deviation and coefficient of variation describe dispersion.

Expected utility incorporates risk preferences, and certainty equivalents express risky prospects as comparable guaranteed amounts. Where probabilities are unavailable, maximin, maximax, minimax regret, Laplace and Hurwicz criteria provide alternative decision rules. Sensitivity analysis identifies critical assumptions, while scenario analysis and stress testing examine coherent combinations of conditions.

Bayesian updating incorporates new evidence, and EVPI and EVSI determine how much information may rationally be worth. Flexible and staged decisions can create real-option value. Quantitative results must nevertheless be evaluated alongside behavioral biases, organizational resilience, worker consequences, legal obligations and model governance.

Key Terms

Ambiguity: A condition in which relevant outcomes or probabilities are unclear.

Bayesian updating: Revision of prior probabilities using new evidence.

Certainty equivalent: Guaranteed payoff with the same utility as a risky prospect.

Coefficient of variation: Standard deviation divided by expected payoff.

Decision tree: Diagram representing sequential decisions, chance events and outcomes.

Expected monetary value: Probability-weighted average monetary payoff.

Expected utility: Probability-weighted average utility of possible outcomes.

Expected value of perfect information: Maximum value of knowing the future state before deciding.

Expected value of sample information: Value of imperfect information before its acquisition cost.

Hurwicz criterion: Uncertainty rule combining best and worst payoffs using an optimism coefficient.

Maximin criterion: Rule selecting the alternative with the highest minimum payoff.

Minimax regret: Rule minimizing the largest opportunity loss.

Real option: Right, but not obligation, to take a future managerial action.

Risk: A condition in which outcomes are uncertain but probabilities can be estimated.

Risk aversion: Preference for less uncertainty when expected payoffs are equal.

Risk premium: Expected payoff minus certainty equivalent.

Scenario analysis: Evaluation of alternatives under coherent combinations of assumptions.

Sensitivity analysis: Examination of how results change when inputs vary.

Standard deviation: Square root of variance, measured in the payoff’s original unit.

Uncertainty: A condition in which reliable probabilities are unavailable or incomplete.

Review Questions

Short Answer Questions

1. Distinguish a decision under risk from a decision under uncertainty.
2. Why is standard deviation easier to interpret than variance?
3. Under what conditions may the coefficient of variation be misleading?
4. What does the certainty equivalent represent?
5. Why can EVPI never be negative?

Long Answer Questions

1. Explain the elements of a managerial decision and illustrate them with an HR example.
2. Compare expected monetary value and expected utility as decision criteria.
3. Critically compare maximin, maximax, minimax regret, Laplace and Hurwicz criteria.
4. Explain how decision trees and Bayesian updating support sequential managerial decisions.
5. Discuss the ethical and governance risks associated with AI-based workforce forecasting.

Analytical / Application Questions

1. A training program produces net benefits of $140,000, $60,000 and −$30,000 with probabilities 0.30, 0.50 and 0.20. Calculate its expected payoff, variance and standard deviation.
2. Using the recruitment campaign data, determine the value of k at which the two campaigns have equal values under RAV = μ − kσ. Interpret the result.
3. Develop a decision tree for permanent hiring, outsourcing and staged hiring under high and low demand. Explain which additional information would be most valuable.

Multiple Choice Questions with Answers

Question 1

Which condition best describes a managerial decision under risk?

A. Every action has a known outcome.

B. Possible outcomes exist and defensible probabilities can be assigned.

C. No possible outcomes can be identified.

D. All alternatives produce the same payoff.

Correct Answer: B. Possible outcomes exist and defensible probabilities can be assigned.

Question 2

The small recruitment campaign has an expected payoff of $60,500 and a standard deviation of approximately $24,520. What is its coefficient of variation?

A. 0.2471

B. 0.4053

C. 0.7716

D. 2.4673

Correct Answer: B. 0.4053

Question 3

What is the variance of the large recruitment campaign presented in the chapter?

A. $601,250,000

B. $2,000,000,000

C. $5,954,000,000

D. $13,918,750,000

Correct Answer: C. $5,954,000,000

Question 4

An HR platform produces gross expected benefit 150,000p + 75,000q and costs $60,000. When does it have positive expected net value?

A. 150,000p + 75,000q > 60,000

B. 150,000p + 75,000q < 60,000

C. 75,000p + 150,000q = 60,000

D. p + q > 60,000

Correct Answer: A. 150,000p + 75,000q > 60,000

Question 5

Which decision rule selects the alternative with the best worst-case payoff?

A. Maximax

B. Laplace

C. Maximin

D. Minimax regret

Correct Answer: C. Maximin

Question 6

Northstar’s permanent-team payoff distribution has which approximate standard deviation?

A. $44,721

B. $77,162

C. $117,978

D. $180,000

Correct Answer: C. $117,978

Question 7

Which statement about the expected value of perfect information is correct?

A. It may be negative when research is expensive.

B. It equals the variance of the preferred alternative.

C. It is the maximum rational price for perfect information under the stated assumptions.

D. It must equal the expected value of sample information.

Correct Answer: C. It is the maximum rational price for perfect information under the stated assumptions.

Question 8

For a risk-averse decision maker with increasing wealth, which conditions characterize the utility function?

A. U′(x) < 0 and U″(x) > 0

B. U′(x) > 0 and U″(x) < 0

C. U′(x) = 0 and U″(x) = 0

D. U′(x) < 0 and U″(x) < 0

Correct Answer: B. U′(x) > 0 and U″(x) < 0

Question 9

Which technique changes several related assumptions together to represent a coherent future environment?

A. Scenario analysis

B. Simple random sampling

C. Minimax regret

D. Variance decomposition

Correct Answer: A. Scenario analysis

Question 10

Using RAV = μ − 0.5σ, which Northstar alternative has the higher risk-adjusted value?

A. Permanent team, approximately $180,000

B. Permanent team, approximately $121,011

C. Contingent workforce, approximately $142,639

D. Contingent workforce, approximately $165,000

Correct Answer: C. Contingent workforce, approximately $142,639