Probability and Business Distributions

Introduction

International business decisions are made under uncertainty. A retailer forecasts demand across countries, a bank evaluates whether a borrower will default, an insurer prices uncertain claims, and a manufacturer monitors defective output. Probability provides a disciplined language for describing uncertainty; probability distributions connect that language to measurable business outcomes.

This chapter develops probability rules, conditional probability, Bayes’ theorem, expected value, and important discrete and continuous distributions. Applications include demand planning, quality control, credit assessment, insurance, and market risk. The central managerial question is not simply “What is likely?” but also “How uncertain is the outcome, what information changes the assessment, and what decision is justified by the expected consequences?”

Learning objectives

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

1. Define experiments, outcomes, events, probabilities, and random variables.
2. Apply addition, multiplication, complement, conditional-probability, and Bayes rules.
3. Calculate and interpret expected value, variance, and standard deviation.
4. Select and use binomial, Poisson, uniform, normal, and exponential models.
5. Analyze business problems involving demand, quality, credit, insurance, and market risk.
6. Recognize assumptions, limitations, ethical risks, and governance requirements in probabilistic analysis.

1. Foundations of probability

A random experiment is a process with an uncertain result. Its sample space, denoted S, is the set of all possible outcomes. An event is a subset of S. For example, in a quality inspection, an outcome may identify whether each item is defective; the event “at least one item is defective” contains all outcomes satisfying that condition.

A probability P(A) assigns a number from 0 to 1 to event A. Probability 0 means impossibility and probability 1 means certainty. For mutually exclusive outcomes that exhaust the sample space, probabilities sum to 1. A probability model should therefore be complete, non-negative, and internally consistent.

Basic rules

The complement of A, written Aᶜ, is the event that A does not occur:

For any events A and B, the general addition rule is:

Here, ∪ means “A or B, or both,” and ∩ means “A and B.” If A and B are mutually exclusive, P(A ∩ B) = 0, so the rule simplifies to P(A ∪ B) = P(A) + P(B).

Conditional probability measures the probability of A when B is known to have occurred:

Rearranging gives the multiplication rule:

Events A and B are independent when knowledge of one does not change the probability of the other. Thus, for independent events:

Independence is an assumption, not a synonym for events being different. For example, demand shocks in two countries may be correlated because both respond to the same exchange-rate movement.

Worked example: supplier quality

A firm buys 60% of its components from Supplier 1 and 40% from Supplier 2. Historical defect rates are 2% and 5%, respectively. Let D denote a defective component. The total probability rule gives:

Therefore:

The overall defect probability is 0.032, or 3.2%.

2. Bayes’ theorem and diagnostic information

Bayes’ theorem updates a prior probability after new evidence is observed:

For mutually exclusive and exhaustive causes A₁, A₂, …, Aₖ:

The terms are interpreted as follows:

The theorem is particularly important where a positive test, warning signal, or observed transaction is not itself conclusive. Base rates matter.

Worked example: credit screening

Suppose 8% of applicants would default within one year. A screening model flags 90% of future defaulters and incorrectly flags 12% of non-defaulters. Let F mean “flagged” and D mean “default.” Then:

The numerator is 0.072. The denominator is 0.072 + 0.1104 = 0.1824. Hence:

Only about 39.5% of flagged applicants are expected to default. A flag should therefore support further assessment rather than automatically justify rejection. The result reflects the relatively low base rate of default and the false-positive rate.

3. Random variables and distributions

A random variable assigns a numerical value to each outcome. A discrete random variable takes countable values, such as the number of defective units. A continuous random variable can take any value in an interval, such as delivery time or daily return.

A probability mass function (PMF) gives probabilities for discrete values:

A cumulative distribution function (CDF) is:

For a continuous variable, probabilities are represented by a probability density function f(x), and:

A density value is not itself a probability; probabilities are areas under the density curve. For a continuous variable, P(X = a) = 0.

Expected value and risk dispersion

For a discrete variable with values x and probabilities p(x), the expected value is:

For a continuous variable with density f(x):

Expected value is a long-run average, not necessarily an outcome that can occur in one transaction. Variance measures dispersion around the mean:

where μ = E(X). Standard deviation is:

Variance is measured in squared units; standard deviation has the same units as X. For a linear transformation Y = a + bX:

Worked example: uncertain project return

A project produces a profit of $100,000 with probability 0.30, $40,000 with probability 0.50, and a loss of $50,000 with probability 0.20. Its expected profit is:

Expected profit is $40,000. However, the possible loss demonstrates why expected value alone is insufficient. A manager should also examine variance, downside probability, liquidity, and risk tolerance.

For a decision with alternatives, expected monetary value (EMV) is the probability-weighted payoff:

where pᵢ is the probability of state i and Vᵢ is its monetary value. EMV is useful for repeated decisions or risk-neutral comparisons; it may not suit a risk-averse firm facing financial distress.

4. Important discrete distributions

Bernoulli distribution

A Bernoulli trial has exactly two outcomes: success, with probability p, and failure, with probability 1 − p. If X = 1 for success and X = 0 for failure:

Its mean and variance are E(X) = p and Var(X) = p(1 − p). Examples include a customer accepting an offer or a loan defaulting.

Binomial distribution

The binomial model counts successes in n independent Bernoulli trials with the same success probability p:

where C(n,x) = n! ÷ [x!(n − x)!]. The mean and variance are:

Conditions include a fixed number of trials, two categories, constant p, and independence. Sampling without replacement from a small finite population may violate independence; a hypergeometric model may then be more appropriate.

Example. If 10% of 20 inspected units are expected to be defective and inspection outcomes are treated as independent, the probability of exactly two defects is:

Poisson distribution

The Poisson model counts events in a fixed interval when events occur independently at a stable average rate λ:

Both mean and variance equal λ. It can model customer arrivals, insurance claims, or order errors, provided the rate is reasonably stable and simultaneous events are rare. If counts are much more variable than their mean, overdispersion suggests that the Poisson assumption is inadequate.

If events occur at rate 4 per hour, the probability of exactly two arrivals in one hour is:

5. Important continuous distributions

Uniform distribution

A continuous variable X is uniform on [a,b] when all values in the interval have equal density:

Its mean and variance are:

For a delivery time uniformly distributed between 2 and 6 days, P(X ≤ 3) = (3 − 2) ÷ (6 − 2) = 0.25.

Normal distribution

The normal distribution is continuous, symmetric, and described by mean μ and standard deviation σ. Its density is:

where σ > 0. Standardization converts X to a standard normal variable Z:

The empirical rule states that approximately 68%, 95%, and 99.7% of observations lie within 1, 2, and 3 standard deviations of μ, respectively, when the distribution is normal. The model is useful for aggregated demand or measurement error, but it can be poor for highly skewed data, bounded quantities, or extreme financial returns.

Example. Daily demand is approximately normal with μ = 500 units and σ = 80 units. Demand above 620 has standardized value:

Using a standard-normal table, P(Z > 1.5) ≈ 0.0668. Thus, demand exceeds 620 units on approximately 6.68% of days. A stock policy based on this result must also consider replenishment time, holding cost, and the cost of shortages.

Exponential distribution

The exponential model describes a waiting time between Poisson events with rate λ:

Its mean is 1 ÷ λ and its variance is 1 ÷ λ². Its survival probability is:

The model has the memoryless property: P(X > s + t | X > s) = P(X > t). This may be useful for simplified service or failure models, but real systems often have aging, maintenance, or time-varying rates.

6. Applications in international business

Demand and inventory

Demand distributions support safety-stock decisions. A normal approximation may be reasonable when demand aggregates many independent influences. If lead-time demand has mean μᴸ and standard deviation σᴸ, a target cycle-service level can be linked to a normal quantile z:

This formula assumes the lead-time demand distribution is approximately normal, parameters are stable, and the selected z corresponds to the desired probability of avoiding a stockout. If demand and lead time are both uncertain, their joint variability must be modeled rather than simply applying a daily-demand formula.

Quality control

Binomial probabilities estimate the number of defective units in a sample. Poisson probabilities can model rare defects per length of material or errors per batch. Managers should distinguish the probability of a defect from the probability that a sampling plan detects it; inspection is evidence, not perfect knowledge.

Credit risk

Credit analysis combines prior default rates, borrower characteristics, and model evidence. A probability of default is not a guarantee. Decisions should incorporate exposure at default, loss given default, collateral, concentration, and stress scenarios. A simplified expected credit loss relationship is:

where PD is probability of default, EAD is exposure at default, and LGD is loss given default as a proportion of exposure. The relationship assumes consistent definitions and compatible time horizons.

Insurance

For claim frequency N and average claim severity C, total loss is:

Expected loss, under independence between frequency and severity, is:

If frequency and severity are dependent, this factorization is not generally valid. Premiums must also cover expenses, capital requirements, uncertainty, and profit—not merely expected claims.

Market risk

Returns are often summarized by mean and standard deviation, but normality can underestimate fat tails and crises. A quantile such as value at risk (VaR) describes a loss threshold at a specified confidence level and horizon; it does not describe losses beyond that threshold. Expected shortfall, the mean loss conditional on exceeding a VaR threshold, provides additional tail information. Both depend heavily on data, horizon, model assumptions, and portfolio composition.

7. Integrated case: international retailer inventory decision

Illustrative case. Meridian Home, a fictional retailer, imports a seasonal product. Historical daily demand is approximately normal with mean 500 and standard deviation 80 units. Supplier lead time is fixed at four days. The firm wants a 95% probability of meeting demand during lead time.

Problem. Determine an approximate reorder point, assuming daily demands are independent.

Analysis. Lead-time demand has mean:

For independent daily demand with standard deviation σ = 80, lead-time variance is 4σ², so:

For a one-sided 95% normal service level, z ≈ 1.645. Safety stock is:

The approximate reorder point is therefore:

Solution and interpretation. Meridian should consider reordering when inventory position reaches approximately 2,263 units, subject to rounding, minimum order quantities, and the distinction between cycle-service level and fill rate. The estimate is not automatically optimal: demand may be seasonal, correlated across days, affected by promotions, or non-normal. Exchange-rate volatility and customs delays may also make lead time uncertain.

Outcome and learning points. The model translates a service target into a measurable buffer, but managerial judgment remains essential. Before implementation, Meridian should back-test the policy, estimate the cost of excess inventory and stockouts, test disruption scenarios, and monitor forecast error. A single historical distribution should not conceal structural changes in markets or supply chains.

8. Limitations, ethics, and governance

Probability models are conditional representations of reality. Common weaknesses include poor data quality, selection bias, changing base rates, dependence incorrectly assumed away, small samples, and confusing correlation with causation. Parameter uncertainty should be distinguished from random variation: even if a binomial model is appropriate, p may itself be uncertain.

In credit, insurance, employment, and pricing, probabilistic systems can reproduce historical discrimination. Governance should include documented variables, explainable decision logic where appropriate, fairness testing, human review, appeal mechanisms, privacy protection, and monitoring for drift. Sensitive attributes and proxies require special scrutiny. A highly accurate model may still be unacceptable if it is unlawful, opaque, or systematically harmful.

AI and machine-learning systems often estimate probabilities, but their outputs require calibration. A group of cases assigned probability 0.20 should experience the event approximately 20% of the time under comparable conditions. Calibration, discrimination, stability, and stress performance should be tested separately. Automated predictions should not be treated as objective facts; they are model-dependent estimates.

Best practice is to:

• define the event, time horizon, population, and unit of analysis clearly;
• validate data and investigate missingness and outliers;
• test independence, distributional form, and parameter stability;
• compare simple benchmarks with complex models;
• report uncertainty, not only point estimates;
• perform sensitivity and stress analysis;
• document assumptions and maintain an audit trail; and
• review models after market, regulatory, or operational change.

Chapter Summary

Probability quantifies uncertain events and supplies rules for combining information. Complements, addition, multiplication, conditional probability, and Bayes’ theorem support operational and strategic decisions. Random variables and distributions describe uncertain numerical outcomes; expected value summarizes long-run average performance, while variance and standard deviation describe dispersion.

The Bernoulli and binomial distributions model binary trials and counts of successes. The Poisson distribution models event counts at a stable rate. Uniform, normal, and exponential distributions model bounded uncertainty, approximately symmetric measurements or aggregated demand, and waiting times, respectively. Each model requires assumptions that should be checked rather than accepted automatically.

Business applications include inventory service levels, quality inspection, credit default assessment, insurance losses, and market risk. Probability is decision support, not certainty. Ethical use requires attention to bias, privacy, explainability, calibration, governance, and changing conditions.

Key Terms

Probability: Numerical measure from 0 to 1 representing uncertainty about an event.

Sample space: Set of all possible outcomes.

Event: Subset of outcomes to which a probability is assigned.

Complement: Event that a specified event does not occur.

Conditional probability: Probability of an event given information about another event.

Independence: Condition in which knowledge of one event does not change the probability of another.

Bayes’ theorem: Rule for updating a prior probability using new evidence.

Random variable: Numerical function assigning values to outcomes.

PMF: Function assigning probabilities to values of a discrete random variable.

CDF: Probability that a random variable is less than or equal to a specified value.

Expected value: Probability-weighted long-run average.

Variance: Expected squared deviation from the mean.

Standard deviation: Square root of variance and measure of dispersion in original units.

Bernoulli distribution: Distribution for one binary trial.

Binomial distribution: Distribution of successes in fixed independent Bernoulli trials.

Poisson distribution: Distribution of event counts over a fixed interval at rate λ.

Normal distribution: Continuous symmetric distribution described by mean and standard deviation.

Exponential distribution: Continuous waiting-time distribution associated with a Poisson process.

Expected loss: Probability-weighted financial loss, often expressed as PD × EAD × LGD.

Calibration: Agreement between predicted probabilities and observed frequencies.

Review Questions

Short Answer Questions

1. Distinguish between mutually exclusive and independent events.
2. Why does Bayes’ theorem require the base rate of an event?
3. What is the difference between a PMF and a probability density function?
4. Interpret expected value and explain why it may not be an attainable outcome.
5. State two assumptions of the binomial distribution.

Long Answer Questions

1. Explain the addition, multiplication, complement, and conditional-probability rules with business examples.
2. Compare the binomial and Poisson distributions, including assumptions, parameters, and applications.
3. Explain how standardization supports probability calculations under the normal distribution.
4. Discuss why expected value alone is insufficient for evaluating an international investment.
5. Evaluate the ethical and governance risks of using probability models in credit decisions.

Analytical / Application Questions

1. A company receives 70% of goods from Supplier A, whose defect rate is 1%, and 30% from Supplier B, whose defect rate is 4%. Calculate the overall defect probability and the probability that a defective item came from Supplier B.
2. Daily demand is normally distributed with mean 1,000 units and standard deviation 150 units. Estimate the probability that demand exceeds 1,300 units using standardization.
3. An insurer expects 3 claims per month under a Poisson model. Calculate the probability of exactly zero claims and explain two reasons why the model might fail in practice.

Multiple Choice Questions with Answers

Question 1

A firm has events A and B with P(A) = 0.40, P(B) = 0.30, and P(A ∩ B) = 0.10. What is P(A ∪ B)?

A. 0.10

B. 0.60

C. 0.70

D. 0.80

Correct Answer: B. 0.60

Question 2

If P(D) = 0.08, P(F | D) = 0.90, and P(F | Dᶜ) = 0.12, what is approximately P(D | F)?

A. 0.08

B. 0.12

C. 0.39

D. 0.90

Correct Answer: C. 0.39

Question 3

Which condition is required for a binomial model?

A. A continuously changing success probability

B. A fixed number of independent trials

C. More than two outcomes per trial

D. Events occurring at a variable rate

Correct Answer: B. A fixed number of independent trials

Question 4

For a Poisson distribution with λ = 4, what are its mean and variance?

A. Mean 2 and variance 4

B. Mean 4 and variance 2

C. Mean 4 and variance 4

D. Mean 16 and variance 4

Correct Answer: C. Mean 4 and variance 4

Question 5

A project yields $50,000 with probability 0.6 and loses $20,000 with probability 0.4. What is its expected monetary value?

A. $22,000

B. $30,000

C. $38,000

D. $70,000

Correct Answer: A. $22,000

Question 6

Daily demand has mean 500 and standard deviation 80. What is the z-score for demand of 620?

A. 0.67

B. 1.00

C. 1.50

D. 2.25

Correct Answer: C. 1.50

Question 7

Which statement about a continuous probability density is correct?

A. The density at a point is always a probability

B. The probability at exactly one point is zero

C. All density values must be less than 0

D. The density must sum rather than integrate to one

Correct Answer: B. The probability at exactly one point is zero

Question 8

If claim frequency and average severity are independent, expected total loss is calculated as:

A. E(N) + E(C)

B. E(N) ÷ E(C)

C. E(N)E(C)

D. E(N) − E(C)

Correct Answer: C. E(N)E(C)

Question 9

In the relationship Expected loss = PD × EAD × LGD, what does LGD represent?

A. Likelihood of geographic disruption

B. Loss given default as a proportion of exposure

C. Long-term growth difference

D. Liability-generating demand

Correct Answer: B. Loss given default as a proportion of exposure

Question 10

A major limitation of using a normal model for extreme market returns is that it may:

A. Have no parameters

B. Underestimate fat-tail risks

C. Allow only nonnegative returns

D. Require discrete observations

Correct Answer: B. Underestimate fat-tail risks