Probability and Decision-Making under Uncertainty

Introduction

Finance is concerned with decisions whose outcomes are not known with certainty. An investor may not know the return on a security, a lender may be uncertain whether a borrower will default, and a treasury manager may not know the future exchange rate or interest rate. Probability provides a disciplined language for representing such uncertainty. It does not eliminate risk; rather, it makes assumptions explicit and allows decisions to be updated when information changes.

This chapter develops the probability concepts most directly useful in finance: probability rules, conditional probability, Bayes’ theorem, random variables, expected value, variance, covariance, correlation, and decision-making under uncertainty. The emphasis is on interpretation as well as calculation. A numerical result is useful only when the events, probabilities, units, and assumptions have been defined correctly.

Learning Objectives

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

1. Define events and assign probabilities consistently.
2. Apply the addition, multiplication, complement, and total-probability rules.
3. Distinguish conditional probability from joint and marginal probability.
4. Use Bayes’ theorem to update beliefs when new evidence arrives.
5. Represent financial outcomes using discrete and continuous random variables.
6. Calculate and interpret expected value, variance, standard deviation, covariance, and correlation.
7. Construct decision trees and compare decisions using expected monetary value.
8. Recognize the limitations of probability models and the governance risks of poor assumptions.

1. Uncertainty, Risk, and Probability

Uncertainty exists when the future outcome is not known. Risk is often used more narrowly to describe uncertainty that can be represented using probabilities. In practice, the distinction is not absolute. A model may assign probabilities to events even when those probabilities are subjective estimates rather than long-run frequencies.

A probability model consists of:

• a sample space, denoted by Ω, containing all possible outcomes;
• events, which are subsets of Ω; and
• a probability measure P that assigns a number from 0 to 1 to each event.

The fundamental probability axioms are:

If A and B cannot occur together, they are mutually exclusive and:

Here, A ∪ B means “A or B,” while A ∩ B means “A and B.” The complement of A, written Aᶜ, means that A does not occur:

Probabilities may be objective, based on observed frequencies, or subjective, based on expert judgement, market information, and economic reasoning. In finance, both forms are common. Historical default frequencies may provide objective evidence, while a portfolio manager’s recession probabilities may be subjective but economically informed.

1.1 Mutually Exclusive and Collectively Exhaustive Events

Events are mutually exclusive if they cannot occur simultaneously. For example, a one-year economic outcome cannot be both “recession” and “expansion” if the categories are defined to exclude one another. Events are collectively exhaustive if they include every possible outcome. A probability table should normally contain mutually exclusive and collectively exhaustive states, with probabilities summing to 1.

The table is a simple discrete probability distribution. Its probabilities are non-negative and sum to one.

2. Rules of Probability

2.1 Addition Rule

For any two events A and B:

The intersection is subtracted because it is counted twice when P(A) and P(B) are added. If A and B are mutually exclusive, P(A ∩ B) = 0, so the rule reduces to the mutually exclusive case.

2.2 Multiplication Rule

For events A and B:

where P(B|A) is the probability of B conditional on A. Equivalently:

These expressions are valid whenever the relevant conditional probabilities are defined. The multiplication rule is particularly important for joint default, joint losses, and sequential decisions.

2.3 Independence

Events A and B are independent if the occurrence of one does not change the probability of the other:

For independent events, the multiplication rule becomes:

Independence is a strong assumption. Two borrowers may appear unrelated but still be exposed to the same recession, commodity shock, or supply-chain disruption. Treating correlated risks as independent can materially understate portfolio risk.

Worked Example 1: Joint Default Probability

Suppose the annual default probability of Firm X is 4% and that of Firm Y is 6%. First assume defaults are independent. The probability that both default is:

Thus, the joint default probability is 0.24%. If a recession makes defaults dependent, the relevant calculation is instead:

If P(Y|X) is estimated at 15%, the joint probability is:

The increase illustrates why unconditional default probabilities alone are insufficient for portfolio credit analysis.

3. Conditional Probability and Bayes’ Theorem

Conditional probability measures the likelihood of one event given that another event has occurred. For P(B) > 0:

The denominator restricts attention to outcomes in which B has occurred. This is why conditional probabilities may differ substantially from unconditional probabilities.

Bayes’ theorem follows from the two forms of the multiplication rule:

If A₁, A₂, …, Aₙ are mutually exclusive and collectively exhaustive causes of B, the denominator can be expanded using the law of total probability:

Therefore:

Bayes’ theorem combines a prior probability, P(A), with evidence, represented by P(B|A), to produce a posterior probability, P(A|B). The prior is the belief before observing the evidence; the posterior is the revised belief after observing it.

Worked Example 2: Updating a Credit-Risk Assessment

A bank classifies a new borrower as high risk with prior probability 10%. A warning signal appears in the borrower’s account. The signal occurs for 70% of high-risk borrowers and for 8% of low-risk borrowers. What is the probability that the borrower is actually high risk given the signal?

Let H denote high risk and S denote the signal. The prior probabilities are P(H) = 0.10 and P(Hᶜ) = 0.90. The likelihoods are P(S|H) = 0.70 and P(S|Hᶜ) = 0.08.

First calculate the total probability of the signal:

Now apply Bayes’ theorem:

The posterior probability is approximately 49.3%. The signal is informative, but it does not imply certainty. A common error is to report 70%, confusing P(S|H) with P(H|S). This is the base-rate problem: the prior frequency of high-risk borrowers matters.

Evidence Quality and Likelihood Ratios

An evidence signal is more useful when it occurs much more often under one hypothesis than another. The likelihood ratio for H against Hᶜ is:

In the example, LR = 0.70 ÷ 0.08 = 8.75. The signal is 8.75 times as likely for a high-risk borrower as for a low-risk borrower. Nevertheless, the posterior also depends on the prior odds. This prevents analysts from treating a strong signal as conclusive without considering the underlying population.

4. Random Variables and Probability Distributions

A random variable assigns a numerical value to each possible outcome. It is commonly denoted by X. A random variable may be:

• Discrete, taking countable values, such as the number of defaults in a portfolio; or
• Continuous, taking values over an interval, such as a daily return or interest rate.

A discrete probability mass function gives P(X = x), and the probabilities must sum to one:

A continuous random variable is described by a probability density function f(x). Probabilities are areas under the density, not the height at a single point:

For a continuous variable, P(X = x) = 0 for any individual point under the usual continuous model. The cumulative distribution function is:

and is applicable to both discrete and continuous variables.

A financial return is often represented as a random variable. If an investment has beginning value V₀ and uncertain ending value V₁, its simple holding-period return is:

This return is dimensionless, usually expressed as a decimal or percentage. The formula assumes no intermediate cash flow. If cash distributions C are received during the period, the holding-period return is:

5. Expected Value and Variability

5.1 Expected Value

For a discrete random variable X with possible values xᵢ and probabilities pᵢ:

The expected value is a probability-weighted average. It is not necessarily an outcome that can actually occur. For a continuous variable:

provided the integral exists.

Worked Example 3: Expected Return

Using the economic-state table, the expected stock return is:

The expected return is 7.0%. This does not mean the investor will necessarily earn 7%; it is the long-run average implied by the specified states and probabilities.

Expected value is linear. For constants a and b:

For random variables X and Y:

The second relationship holds whether or not X and Y are independent.

5.2 Variance and Standard Deviation

Variance measures dispersion around the expected value. For a discrete variable:

An equivalent computational form is:

Standard deviation is the positive square root of variance:

If X is a return measured in percentage points, variance has squared percentage units, while standard deviation has percentage-point units. Standard deviation is therefore easier to interpret as a measure of return volatility.

For the example:

Thus:

The standard deviation is approximately 12.17%, under the stated scenario model.

5.3 Semivariance and Downside Measures

Standard deviation treats positive and negative deviations symmetrically. For investors concerned primarily with losses, downside measures may be more appropriate. A target-semivariance can be written as:

where T is a target return. The expression measures squared shortfalls below T and assigns zero to outcomes at or above T. Other downside measures include shortfall probability, value at risk, and expected shortfall. These measures require careful definition of the horizon, confidence level, and loss convention.

6. Covariance, Correlation, and Portfolio Risk

Covariance measures the direction in which two random variables move together:

For a discrete joint distribution:

Covariance is positive when high values of X tend to occur with high values of Y, and negative when high values of one tend to occur with low values of the other. Its magnitude depends on the units of X and Y, so it is difficult to compare across pairs.

Correlation standardizes covariance:

provided σₓ and σᵧ are positive. Correlation lies between −1 and +1. Zero correlation does not generally imply independence; it implies only absence of linear association. Under joint normality, zero correlation does imply independence, but that is a special assumption.

For a two-asset portfolio with weights w₁ and w₂, returns R₁ and R₂, and w₁ + w₂ = 1:

Expected portfolio return is:

Portfolio variance is:

Using correlation, the covariance term may be written as ρ₁₂σ₁σ₂. The formula shows why diversification depends on co-movement, not merely on the number of securities. The weights must be expressed as proportions of portfolio value, and all returns must refer to the same horizon.

7. Decision-Making under Uncertainty

A decision problem has alternatives, states of nature, probabilities, and payoffs. A decision maker chooses an action before the state is known. A decision tree is useful when decisions and information arrive sequentially.

Expected monetary value (EMV) for action a is:

The risk-neutral rule selects the action with the highest EMV. This rule is appropriate when the decision maker evaluates alternatives by expected monetary payoff alone. It may not be appropriate for a risk-averse investor.

Risk Aversion and Utility

A risk-averse decision maker may maximize expected utility rather than expected wealth:

where U is a utility function. A common concave utility function is U(W) = √W for non-negative wealth W. Concavity means that the marginal utility of wealth decreases as wealth increases. Consequently, a certain payoff may be preferred to a gamble with the same expected payoff.

Certainty equivalent is the guaranteed amount CE that gives the same utility as the risky prospect:

For a risk-averse individual, CE is generally less than the expected monetary payoff. The difference is the risk premium in a broad decision-theoretic sense.

Information Value

The expected value of perfect information (EVPI) measures the maximum amount a decision maker should pay to know the future state before choosing an action:

EVPI cannot be negative. Imperfect information has lower value because it does not reveal the state perfectly. Information should be purchased only when its expected benefit exceeds its cost and when it can be operationally used.

8. Applied Case: A Bank’s Early-Warning Decision

The following is an illustrative case, not a report of a specific bank.

Background

A commercial bank must decide whether to renew a one-year working-capital facility for a corporate borrower. Internal data identify two borrower states: healthy (H) and distressed (D). Before receiving new information, the credit team estimates P(D) = 0.12 and P(H) = 0.88.

The bank’s estimated net payoff from renewal is 2 monetary units if the borrower is healthy and −10 if distressed. If the bank declines renewal, the payoff is 0 in either state because it forgoes the lending opportunity. The bank also receives an early-warning signal, S, from payment behaviour.

The signal is observed before the decision. Its characteristics are P(S|D) = 0.75 and P(S|H) = 0.10.

Problem

Should the bank renew when the warning signal appears, and what is the value of the signal?

Analysis

The probability of a signal is:

Bayes’ theorem gives:

Thus, after the signal, the probability of distress is approximately 50.56%. The expected payoff from renewal after S is:

The bank should decline renewal after the signal because 0 is greater than −4.0672.

If no signal appears, P(Sᶜ) = 0.822. The posterior probability of distress is:

The expected payoff from renewal when no signal appears is:

The bank should renew if no signal appears.

Without the signal, the expected payoff from renewal would be:

With the signal and an optimal response, the expected value is:

Therefore, the expected value of the signal is approximately:

Before considering collection and implementation costs, the signal is worth approximately 0.725 monetary units per facility. The result depends on the accuracy of the signal, the prior distress rate, and the payoff estimates.

Outcome and Learning Points

The signal changes the action rather than merely changing a reported probability. This is the practical purpose of Bayesian analysis. The case also demonstrates that an informative signal may have economic value only when it changes a decision. If the bank would renew or decline regardless of the signal, the signal’s decision value would be zero even if its statistical accuracy were high.

9. Model Risk, Governance, and Ethical Issues

Probability models can fail for reasons that are not computational. Important sources of model risk include:

• Poor event definitions: “Default,” “recession,” and “loss” must be operationally defined.
• Inadequate data: Small or biased samples can produce unstable probabilities.
• Regime change: Historical relationships may fail after regulatory, technological, or macroeconomic changes.
• Dependence assumptions: Correlated defaults and common shocks are often underestimated.
• Probability calibration: Forecast probabilities should be compared with subsequent outcomes.
• Selection bias: Data may exclude borrowers who were rejected, creating a misleading sample.
• Confusing association with causation: A predictive signal need not be a causal explanation.
• Overconfidence: Precise-looking probabilities may conceal substantial estimation uncertainty.

Financial institutions should document model purpose, data sources, assumptions, validation procedures, limitations, and approval responsibilities. Independent validation, stress testing, back-testing, sensitivity analysis, and monitoring of model drift are important controls.

Ethically, probability-based decisions should not rely on prohibited discriminatory characteristics or on proxy variables that reproduce unfair treatment. Automated credit systems should be explainable at an appropriate level, subject to human oversight, and monitored for disparate error rates. A model should support accountable decision-making, not obscure responsibility behind a numerical score.

10. Technology and Contemporary Practice

Modern financial institutions use statistical software, simulation, cloud computing, and machine-learning systems to estimate probabilities and generate scenarios. Monte Carlo simulation repeatedly samples from specified distributions to approximate a distribution of outcomes. Its results are conditional on the input model; a large number of simulations cannot correct an unrealistic dependence structure or an incorrect loss distribution.

Machine-learning systems can improve predictive accuracy, particularly with nonlinear relationships and large datasets. However, they may produce poorly calibrated probabilities, become unstable under distributional change, or be difficult to explain. Best practice is to evaluate discrimination and calibration separately, retain interpretable benchmarks, test out-of-sample performance, and combine automated outputs with governance and expert review.

Best Practices for Financial Probability Analysis

1. Define the event, horizon, population, and unit of analysis before calculating.
2. Ensure probabilities are non-negative and collectively exhaustive probabilities sum to 1.
3. Distinguish P(A|B) from P(B|A); use Bayes’ theorem when reversing conditioning.
4. State whether independence is assumed and justify it economically.
5. Match return horizons, currency units, and compounding conventions.
6. Report both expected outcomes and measures of dispersion or downside risk.
7. Test sensitivity to priors, likelihoods, correlations, and scenario probabilities.
8. Separate model uncertainty from randomness within the model.
9. Validate forecasts against later outcomes and investigate calibration failures.
10. Link information to an actual decision and compare benefits with acquisition costs.

Chapter Summary

Probability provides the formal structure for representing financial uncertainty. The addition rule handles alternative events, while the multiplication rule handles joint events and sequential relationships. Conditional probability changes the relevant reference set, and Bayes’ theorem updates prior beliefs using new evidence. Random variables represent uncertain financial quantities; their distributions describe possible values and associated probabilities.

Expected value is the probability-weighted average outcome, while variance and standard deviation measure dispersion. Covariance and correlation describe co-movement and are central to portfolio diversification. Expected monetary value supports risk-neutral choices, whereas expected utility incorporates risk aversion. Decision trees and information-value analysis show how new evidence can change actions and create economic value.

Financial probability models are only as reliable as their definitions, data, assumptions, and governance. Dependence, regime change, estimation error, selection bias, and ethical risks must be considered. Sound practice combines mathematical calculation with economic reasoning, sensitivity analysis, validation, and accountable professional judgement.

Key Terms

Probability: A number from 0 to 1 representing the likelihood of an event.

Sample space: The set of all possible outcomes.

Event: A subset of the sample space.

Complement: The event that A does not occur, written Aᶜ.

Mutually exclusive events: Events that cannot occur simultaneously.

Conditional probability: The probability of one event given another event.

Joint probability: The probability that two events occur together.

Independence: A condition in which one event does not alter the probability of another.

Bayes’ theorem: A rule for updating a probability using new evidence.

Prior probability: Belief assigned before observing new evidence.

Posterior probability: Updated belief after observing evidence.

Random variable: A numerical representation of an uncertain outcome.

Probability distribution: A description of possible values and their probabilities.

Expected value: The probability-weighted average of a random variable.

Variance: The expected squared deviation from the mean.

Standard deviation: The square root of variance.

Covariance: A measure of joint directional movement between two variables.

Correlation: Standardized covariance ranging from −1 to +1.

Expected monetary value: The probability-weighted payoff of a decision.

Expected utility: The probability-weighted utility of outcomes, incorporating preferences toward risk.

Review Questions

5 Short Answer Questions

1. What is the difference between a mutually exclusive pair of events and an independent pair of events?
2. Why is P(A|B) generally different from P(B|A)?
3. What does the expected value of a financial return represent?
4. Why is covariance important in portfolio analysis?
5. What is the purpose of an independent model-validation process?

5 Long Answer Questions

1. Explain the addition and multiplication rules of probability, including the conditions under which each simplified form applies.
2. Derive Bayes’ theorem from the multiplication rule and explain the roles of prior, likelihood, evidence, and posterior probability.
3. Discuss why standard deviation alone may be insufficient for an investor concerned primarily with downside losses.
4. Explain how correlation affects the risk of a two-asset portfolio and why diversification does not require negative correlation.
5. Compare expected monetary value and expected utility as decision criteria, including the role of risk aversion.

3 Analytical / Application Questions

1. A portfolio manager estimates a 25% probability of recession. A bond has a return of −8% in recession and 6% otherwise. Calculate its expected return and explain why the answer is not a forecast of the actual realized return.
2. A fraud signal occurs for 80% of fraudulent transactions and 5% of legitimate transactions. If 2% of transactions are fraudulent, calculate the probability that a transaction is fraudulent given the signal. Interpret the base-rate effect.
3. Two assets have standard deviations of 12% and 20%, portfolio weights of 40% and 60%, and correlation of −0.25. Explain qualitatively how the negative correlation affects portfolio risk, then calculate the portfolio variance using the two-asset variance formula.

Multiple Choice Questions with Answers

Question 1

If P(A) = 0.30, P(B) = 0.40, and A and B are independent, what is P(A ∩ B)?

A. 0.10

B. 0.12

C. 0.30

D. 0.70

Correct Answer: B. 0.12

Question 2

Which expression correctly represents Bayes’ theorem for P(A|B)?

A. P(A)P(B)

B. P(A ∪ B) ÷ P(B)

C. [P(B|A)P(A)] ÷ P(B)

D. P(B|A) ÷ P(A)

Correct Answer: C. [P(B|A)P(A)] ÷ P(B)

Question 3

A share returns −10% with probability 0.25 and 20% with probability 0.75. What is its expected return?

A. 5.0%

B. 10.0%

C. 12.5%

D. 15.0%

Correct Answer: B. 10.0%

Question 4

Which statement about correlation is correct?

A. It must lie between 0 and 1.

B. It measures only the expected return of an asset.

C. It is standardized covariance and lies between −1 and +1.

D. Zero correlation always proves independence.

Correct Answer: C. It is standardized covariance and lies between −1 and +1.

Question 5

A signal occurs in 70% of distressed borrowers and 10% of healthy borrowers. If the prior probability of distress is 20%, what is P(signal)?

A. 0.14

B. 0.16

C. 0.22

D. 0.80

Correct Answer: C. 0.22

Question 6

For a risk-averse investor, which relationship generally holds between certainty equivalent and expected monetary payoff?

A. Certainty equivalent is greater.

B. Certainty equivalent is equal for every gamble.

C. Certainty equivalent is generally lower.

D. There is no relationship.

Correct Answer: C. Certainty equivalent is generally lower.

Question 7

Which assumption is required for the simplified formula P(A ∩ B) = P(A)P(B)?

A. A and B are mutually exclusive.

B. A and B are independent.

C. A and B are collectively exhaustive.

D. P(A) equals P(B).

Correct Answer: B. A and B are independent.

Question 8

If two assets have positive covariance, what does this imply most directly?

A. They always have positive returns.

B. They tend to move in the same direction relative to their means.

C. They have identical expected returns.

D. They are independent.

Correct Answer: B. They tend to move in the same direction relative to their means.

Question 9

Expected monetary value is most directly calculated by:

A. Selecting the largest possible payoff.

B. Averaging payoffs without probabilities.

C. Summing each payoff multiplied by its state probability.

D. Subtracting variance from every payoff.

Correct Answer: C. Summing each payoff multiplied by its state probability.

Question 10

What is the main purpose of the posterior probability in Bayesian analysis?

A. To describe beliefs before evidence is observed.

B. To update beliefs after evidence is observed.

C. To guarantee the evidence is causal.

D. To eliminate all model uncertainty.

Correct Answer: B. To update beliefs after evidence is observed.