Decision Analysis, Risk and Simulation

Introduction

Business analytics rarely produces decisions under complete certainty. A manager may choose a supplier before demand is known, launch a product before competitors respond, or invest in technology before its benefits are established. Decision analysis provides a disciplined way to represent alternatives, uncertain events, consequences, and preferences. It does not eliminate uncertainty; it makes assumptions visible and compares choices consistently.

This chapter develops a practical framework for constructing payoff tables and decision trees, calculating expected monetary value (EMV), incorporating risk preferences through utility, valuing information, conducting scenario and sensitivity analysis, and using Monte Carlo simulation. The methods apply to pricing, capacity planning, inventory, marketing, finance, project selection, and operations.

Learning objectives

By the end of this chapter, you should be able to:

1. Distinguish decisions, states of nature, outcomes, risk, and uncertainty.
2. Construct and interpret payoff tables and decision trees.
3. Calculate EMV, expected opportunity loss, expected value of perfect information, and expected value of sample information.
4. Explain risk attitudes and use utility functions appropriately.
5. Perform scenario, break-even, and sensitivity analysis.
6. Design and interpret a Monte Carlo simulation for an uncertain business problem.
7. Evaluate assumptions, communicate limitations, and govern analytical decisions responsibly.

1. Foundations of decision analysis

A decision problem contains four elements:

• Decision alternatives: actions controlled by the decision-maker.
• States of nature: uncertain conditions not controlled by the decision-maker.
• Probabilities: beliefs about the likelihood of states, when probabilities can reasonably be assigned.
• Payoffs: outcomes such as profit, cost, revenue, service level, or customer value.

A decision under certainty has one known state. A decision under risk has several possible states with known or estimated probabilities. A decision under uncertainty has several possible states but probabilities are unavailable, disputed, or too unreliable to use confidently.

A probability model should be based on historical data, experiments, market research, expert elicitation, or a transparent combination of these sources. Probabilities are not objective facts in every business setting; they are often conditional beliefs based on information available at the time of decision.

Payoff tables

A payoff table lists the payoff from each action under each possible state.

The entries must be comparable: they should use the same time period, currency, and definition of payoff. If alternatives have different lifetimes, payoffs should normally be converted to present values. If costs and benefits occur at different times, discounting may be required.

2. Expected monetary value

Expected monetary value is the probability-weighted average payoff of an alternative. For alternative a and states s:

Here, p(s) is the probability of state s, and V(a,s) is the payoff from alternative a in that state. The probabilities must be non-negative and sum to 1. EMV is appropriate when repeated decisions are made under similar conditions, or when the decision-maker is approximately risk neutral.

For the table above:

• Small facility: 0.40 × £80,000 + 0.60 × £120,000 = £104,000
• Large facility: 0.40 × £20,000 + 0.60 × £220,000 = £140,000
• Do not invest: £0

The EMV recommendation is therefore the large facility. This does not mean that the large facility is certain to be best. It has a £20,000 payoff in the low-demand state, whereas the small facility produces £80,000. EMV summarizes uncertainty; it does not remove it.

Expected opportunity loss

Opportunity loss, or regret, is the difference between the best payoff available in a state and the payoff from the selected alternative:

The expected opportunity loss is:

In the example, the best payoff under low demand is £80,000 and under high demand is £220,000. The regret table is:

The alternative with the highest EMV also has the lowest EOL here. This relationship is common when the same probabilities and payoffs are used, but EOL should not be treated as a separate decision rule without understanding its assumptions.

3. Decision trees and sequential decisions

A decision tree is useful when decisions occur in stages or when later actions depend on information received earlier. A square node represents a decision; a circle represents a chance event; terminal branches show payoffs. The tree is solved from right to left, a process called rollback.

Figure: A basic decision tree for a facility decision.

At a chance node, calculate the probability-weighted value of outgoing branches. At a decision node, select the branch with the greatest value, subject to the organization’s objective and constraints. A tree must not double-count probabilities or omit costs incurred before later events.

Sequential information example

Suppose a firm can first conduct a market study costing £15,000. The study predicts favourable or unfavourable demand, but predictions are imperfect. The expected value of the study is calculated by:

1. Assigning probabilities to study results.
2. Updating demand probabilities after each result using Bayes’ rule.
3. Choosing the best action after each result.
4. Calculating the expected payoff of the study.
5. Subtracting the study cost.

Bayes’ rule is:

Here, S is a state of demand and R is an observed research result. The formula applies when the conditional probabilities are defined consistently and P(R) > 0.

4. Utility and risk preference

EMV treats one pound of gain as having the same decision significance wherever it occurs. People and organizations may instead be risk averse, risk neutral, or risk seeking.

• A risk-neutral decision-maker compares EMV.
• A risk-averse decision-maker prefers a more certain outcome when two options have the same EMV.
• A risk-seeking decision-maker may prefer variability.

Expected utility is:

U(V) is the utility assigned to payoff V. The alternative with the greatest expected utility is preferred. A utility function must reflect the relevant decision-maker’s preferences and should not be presented as an objective replacement for judgment.

A commonly used risk-averse function is exponential utility:

where x is monetary payoff and R is a positive risk-tolerance parameter in the same monetary units as x. Larger R represents lower risk aversion. This function is useful for modelling but requires careful calibration. Utility can also incorporate non-monetary consequences such as safety, reputation, or environmental impact.

The certainty equivalent is the guaranteed amount producing the same utility as a risky prospect. It satisfies:

For a risk-averse decision-maker, the certainty equivalent is normally less than EMV. The difference, EMV − CE, is the risk premium.

5. Value of information

Information has value only if it changes a decision or improves its consequences. The expected value of perfect information (EVPI) is the maximum amount worth paying for complete knowledge of the future state:

Equivalently, EVPI equals the minimum expected opportunity loss across alternatives. It is an upper bound: no imperfect forecast should cost more than EVPI in a strictly monetary model.

The expected value of sample information (EVSI) measures the value of an imperfect study:

If the study costs C, its net value is EVSI − C. The study is economically attractive only when its value exceeds its cost, after considering implementation time, bias, and operational disruption.

6. Scenario and sensitivity analysis

Scenario analysis examines coherent combinations of assumptions. For example:

Scenarios are not necessarily probabilities. They are structured narratives used to test strategy. If probabilities are assigned, they must be justified and should not be confused with forecasts.

Sensitivity analysis changes one input at a time, or a defined group of inputs, to identify drivers of the decision. A break-even probability can be calculated for two alternatives. For alternatives A and B with two states, A is preferred when:

Solving for p gives the threshold probability at which the preferred alternative changes, provided the coefficient of p is non-zero. Sensitivity analysis should also consider correlated inputs. Changing demand and price independently may create unrealistic combinations if they are economically related.

Useful outputs include tornado charts, threshold tables, spider plots, and probability-of-loss measures. A model is decision-robust when its recommendation remains stable across a credible range of assumptions.

7. Monte Carlo simulation

Monte Carlo simulation represents uncertain inputs with probability distributions, repeatedly samples values, calculates an outcome, and summarizes the resulting distribution. Unlike EMV, which often reports one average, simulation reveals dispersion, downside risk, percentiles, and the probability of breaching a target.

A general simulation procedure is:

1. Define the outcome measure and decision alternatives.
2. Identify uncertain inputs and fixed inputs.
3. Select distributions using data and domain knowledge.
4. Specify dependencies among inputs.
5. Generate random draws for each trial.
6. Calculate the outcome using the business model.
7. Repeat for many trials.
8. Summarize and validate the output.

For profit:

For a simple sales model, if demand D, price P, variable cost per unit c, and fixed cost F are uncertain:

All monetary inputs must use the same currency and time basis. A distribution should represent plausible values, not merely produce computational variety. Common choices include normal, triangular, uniform, lognormal, binomial, and empirical distributions. A triangular distribution is often useful when experts can specify minimum, most likely, and maximum values, but it can falsely imply linear density if those assumptions are not credible.

Simulation output

Important summaries include:

• Mean and median profit.
• Standard deviation, where meaningful.
• Percentiles such as P5, P50, and P95.
• Probability of loss: P(Profit < 0).
• Probability of meeting a target: P(Profit ≥ target).
• Minimum and maximum observed values, interpreted cautiously.

The sample mean after n trials is:

The simulation standard error of the mean is approximately:

where s is the sample standard deviation. More trials reduce Monte Carlo sampling error, but they do not correct a biased model or poor distributions. Random seeds support reproducibility; independent replications can help assess simulation stability.

Illustrative simulation design

A retailer considers an order quantity. Demand is modelled as triangular with minimum 800, most likely 1,200, and maximum 2,000 units. Selling price is £30, unit purchase cost is £12, and unsold units have a salvage value of £5. For order quantity Q, sales are min(D,Q), and leftover inventory is max(Q−D,0). Profit is:

The analyst simulates several candidate values of Q. The chosen order quantity should not be based only on mean profit. A quantity with slightly higher mean profit but a 25% probability of loss may be less suitable than one with lower mean profit and a 5% probability of loss, depending on cash constraints and risk preference.

Applied case: capacity planning for a meal-delivery service

This is an illustrative case. A meal-delivery company must choose between a standard kitchen and an expanded kitchen for the next year. Demand may be low, medium, or high. The finance team estimates probabilities of 0.30, 0.45, and 0.25. Net payoffs, after operating costs, are:

Analysis

Standard EMV is 0.30×90,000 + 0.45×150,000 + 0.25×175,000 = £139,250. Expanded EMV is 0.30×20,000 + 0.45×160,000 + 0.25×280,000 = £133,000. Under risk neutrality, the standard kitchen is preferred.

However, the expanded kitchen produces the highest payoff in the high-demand state and may create strategic capacity. The team should therefore conduct sensitivity analysis on the probability of high demand, test demand-price dependence, and simulate monthly demand rather than treating annual demand as a single event. It should also assess labour availability, service quality, food waste, and the reversibility of the investment.

Outcome and learning points

The appropriate recommendation is not automatically “choose the highest EMV.” The decision should combine EMV with downside probability, utility, capacity constraints, and information quality. A pilot campaign or customer survey may have positive EVSI if it materially improves demand estimates and costs less than the information value.

The case demonstrates that:

• A payoff table is a useful first representation.
• EMV can favour a safer alternative even when another has greater upside.
• A simulation can reveal monthly variability hidden by annual averages.
• Non-financial effects and strategic flexibility belong in the decision model.

8. Professional practice, limitations and governance

Decision models are conditional arguments: if the assumptions hold, the recommendation follows. They are not proof of what will happen. Major limitations include poor data, unstable markets, omitted alternatives, dependent uncertainties, probability miscalibration, and false precision.

Best practice includes:

• Documenting data sources, assumptions, dates, and units.
• Separating observed data from expert judgments.
• Testing alternative distributions and correlations.
• Reporting ranges and probabilities rather than only point estimates.
• Performing validation and independent review.
• Keeping a record of model versions and random seeds.
• Explaining results in business language without overstating certainty.
• Including operational, legal, safety, privacy, equality, and environmental consequences.

Technology can automate simulation, sensitivity charts, and dashboard reporting. AI systems can assist with scenario generation or pattern detection, but generated assumptions require human validation. Models should not use personal or commercially sensitive data without lawful authority, appropriate security, and purpose limitation. A decision should remain auditable: stakeholders must be able to understand the inputs, logic, uncertainty, and responsibility for action.

Chapter Summary

Decision analysis structures choices under certainty, risk, and uncertainty. Payoff tables compare alternatives across states of nature, while decision trees represent sequential decisions and information. EMV is the probability-weighted payoff and is most appropriate for risk-neutral or repeated-decision settings. Expected opportunity loss measures regret. Utility analysis incorporates risk preference, and the certainty equivalent translates a risky prospect into an equivalent guaranteed payoff.

EVPI sets an upper bound on the value of perfect information, while EVSI evaluates imperfect research after accounting for study cost. Scenario analysis tests coherent futures; sensitivity analysis identifies assumptions that can change a recommendation. Monte Carlo simulation repeatedly samples uncertain inputs to produce a distribution of outcomes, allowing analysts to estimate percentiles and probabilities of loss or target achievement. Sound practice requires realistic distributions, dependence modelling, validation, transparent documentation, and ethical governance.

Key Terms

Decision alternative: An action available to the decision-maker.

State of nature: An uncertain condition outside the decision-maker’s control.

Payoff: The consequence of an alternative under a particular state.

Payoff table: A matrix of alternatives, states, and resulting payoffs.

Decision tree: A graphical representation of decisions, chance events, and outcomes.

Expected monetary value: The probability-weighted average payoff of an alternative.

Opportunity loss: The payoff forgone by not choosing the best alternative for a state.

Expected opportunity loss: The probability-weighted average opportunity loss.

Risk: Uncertainty represented by credible probabilities.

Uncertainty: Situations in which probabilities are unavailable or unreliable.

Utility: A numerical representation of preference for outcomes.

Risk aversion: Preference for lower variability, holding expected value constant.

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

EVPI: Expected value of perfect information.

EVSI: Expected value of sample information.

Scenario analysis: Evaluation of coherent combinations of assumptions.

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

Monte Carlo simulation: Repeated random sampling to estimate an outcome distribution.

Percentile: A value below which a specified proportion of simulated outcomes falls.

Model governance: Controls ensuring that an analytical model is valid, documented, fair, and accountable.

Review Questions

Short Answer Questions

1. What are the four basic elements of a decision problem?
2. When is EMV an appropriate decision criterion?
3. What is the difference between a decision node and a chance node?
4. Define EVPI and explain why it is an upper bound on information cost.
5. Why can a Monte Carlo simulation be misleading even when it uses many trials?

Long Answer Questions

1. Explain how to construct and solve a decision tree using rollback, including the treatment of decision and chance nodes.
2. Compare EMV, expected utility, and certainty equivalent as methods for making decisions under risk.
3. Explain how EVSI can be calculated for a market research study and identify the assumptions required.
4. Discuss the differences between scenario analysis and sensitivity analysis, including suitable business applications.
5. Describe a complete Monte Carlo simulation workflow from problem definition through communication of results.

Analytical / Application Questions

1. A firm has two alternatives with payoffs of £40,000 and £100,000 under low and high demand for Alternative A, and £10,000 and £160,000 for Alternative B. If high demand has probability 0.60, calculate both EMVs and identify the preferred alternative.
2. Design a simulation model for a retailer choosing inventory quantity when demand, price, and salvage value are uncertain. Specify inputs, an outcome equation, at least three outputs, and two dependency or validation checks.
3. A manager prefers a stable project with a lower EMV to a volatile project with a higher EMV. Explain how utility, certainty equivalent, sensitivity analysis, and downside probabilities could be used to justify or challenge that preference.

Multiple Choice Questions with Answers

Question 1

Which expression correctly defines expected monetary value?

A. The largest payoff across all states

B. The probability-weighted sum of payoffs

C. The smallest opportunity loss in one state

D. The average of probabilities only

Correct Answer: B. The probability-weighted sum of payoffs

Question 2

In a decision tree, a circle normally represents:

A. A decision node

B. A fixed cost

C. A chance event

D. A terminal payoff

Correct Answer: C. A chance event

Question 3

Using the facility example, what is the EMV of the large facility?

A. £24,000

B. £104,000

C. £140,000

D. £220,000

Correct Answer: C. £140,000

Question 4

Expected opportunity loss measures:

A. The amount paid for research

B. The difference between the best state payoff and the selected payoff

C. The variance of a payoff distribution

D. The probability of a high-demand state

Correct Answer: B. The difference between the best state payoff and the selected payoff

Question 5

A risk-averse decision-maker generally prefers:

A. Greater variability at the same EMV

B. A guaranteed outcome with lower utility

C. Less variability when expected value is otherwise comparable

D. Only the alternative with the highest maximum payoff

Correct Answer: C. Less variability when expected value is otherwise comparable

Question 6

EVPI is best described as:

A. The cost of every possible survey

B. The value of complete knowledge of the future state before deciding

C. The expected value of an inaccurate forecast

D. The difference between revenue and cost

Correct Answer: B. The value of complete knowledge of the future state before deciding

Question 7

Which activity is scenario analysis?

A. Changing one parameter by 1% while holding all others fixed

B. Creating coherent downside, base, and upside futures

C. Repeating random draws thousands of times

D. Selecting the largest historical observation

Correct Answer: B. Creating coherent downside, base, and upside futures

Question 8

For profit = demand × (price − unit cost) − fixed cost, all monetary inputs should:

A. Use consistent currency and time basis

B. Be expressed as probabilities

C. Be independent by definition

D. Be rounded to whole millions

Correct Answer: A. Use consistent currency and time basis

Question 9

Which output is particularly useful for assessing downside risk in a simulation?

A. The maximum number of trials

B. The probability that profit is below zero

C. The random seed alone

D. The number of model variables

Correct Answer: B. The probability that profit is below zero

Question 10

What does increasing the number of simulation trials primarily reduce?

A. Model bias from an incorrect distribution

B. The real-world business risk

C. Monte Carlo sampling error

D. Dependence between inputs

Correct Answer: C. Monte Carlo sampling error