What Is a Probability Distribution?
A probability distribution describes how the probabilities of all possible outcomes for a random variable are allocated. It tells you, for each potential outcome or range of outcomes, how likely that outcome is to occur. Graphically, distributions may appear as a list of probabilities (discrete case) or as a smooth curve (continuous case).
Key takeaways
– A valid probability distribution assigns nonnegative probabilities and sums (or integrates) to 1.
– Discrete distributions apply when outcomes are countable (e.g., number of defaults); continuous distributions apply when outcomes vary over intervals (e.g., returns).
– Common distributions in finance: normal, lognormal, binomial, Poisson, Student’s t, exponential, beta.
– Finance models often assume normality, but asset returns commonly exhibit fat tails (excess kurtosis) and skewness—important for risk management.
– The central limit theorem explains why averages of independent samples tend toward a normal distribution even if the underlying data are not normal.
Types of probability distributions
– Discrete: Values are countable (finite or countably infinite). Examples: Bernoulli, binomial, Poisson, geometric.
– Continuous: Values lie on a continuum; probabilities are described by a probability density function (PDF). Examples: normal, lognormal, exponential, Student’s t, beta.
Important distributions (brief)
– Binomial (discrete): Models the number of successes in n independent trials with success probability p. Useful for yes/no outcomes over fixed trials.
– Normal (continuous): Symmetric, bell-shaped; fully described by mean µ and standard deviation σ. Common but may understate tail events in finance.
– Lognormal (continuous): If log(price) ~ normal, then price is lognormal. Stock prices (nonnegative, potentially unbounded upward) are often modeled as lognormal.
– Poisson (discrete): Models the count of rare events in a fixed interval with mean λ. Useful for arrival counts (defaults, claims).
– Student’s t: Similar to normal but with heavier tails—better for modeling returns with more extreme moves.
What makes a probability distribution valid?
– Nonnegativity: P(X = x) ≥ 0 (discrete) or f(x) ≥ 0 (continuous) for all x.
– Total probability = 1: sum over discrete outcomes = 1, or integral of PDF over support = 1.
– Support is specified: define the possible values (e.g., [0, ∞) for exponential, all real numbers for normal).
Probability distribution vs. cumulative distribution
– PDF (continuous) / PMF (discrete): gives density or point probabilities.
– CDF: gives P(X ≤ x). It is nondecreasing, starts at 0 and approaches 1.
Probability vs. odds
– Probability p is the chance an event occurs (0 ≤ p ≤ 1).
– Odds (in favor) = p / (1 − p). Conversely, p = odds / (1 + odds).
– Example: p = 0.25 → odds = 0.25 / 0.75 = 1/3 (or “1 to 3” against).
Law of large numbers (LLN)
– As the number of independent, identically distributed trials increases, the sample average converges to the population expected value. LLN underpins why long-run averages become stable.
Central limit theorem (CLT)
– For a sufficiently large sample size, the distribution of the sample mean is approximately normal with mean µ and variance σ^2/n, regardless of the original distribution (provided finite variance). CLT justifies many inferential procedures and the ubiquity of normal-based approximations.
Example calculations (simple, practical)
1) Binomial probability
– Problem: Probability of exactly 7 heads in 10 fair coin flips.
– Formula: P(k) = C(n,k) p^k (1−p)^(n−k)
– Calculation: C(10,7)·0.5^7·0.5^3 = 120·0.5^10 ≈ 0.1172 (≈11.72%).
2) Poisson probability
– Problem: Average λ = 2 emails/hour. Probability of exactly 5 emails in an hour:
– P(k) = e^{−λ} λ^k / k!
– Calculation: e^{−2}·2^5/5! ≈ 0.0361 (≈3.61%).
3) Normal probability and VaR example (finance)
– Suppose daily portfolio returns have µ = 0.05% (0.0005) and σ = 2% (0.02).
– Probability daily loss worse than −5%: z = (−0.05 − 0.0005)/0.02 ≈ −2.525 → P ≈ 0.0058 (0.58%).
– 95% daily VaR (normal assumption): z_{0.05} ≈ −1.645 → VaR = µ + zσ ≈ 0.0005 − 1.645·0.02 ≈ −0.03235 → expect loss ≥ 3.235% on the worst 5% of days.
Types of distributions commonly used in finance
– Normal: returns (often assumed), risk models.
– Lognormal: asset prices (positive support).
– Student’s t: returns with fat tails.
– Binomial: option pricing (binomial tree), credit event counts.
– Poisson: modeling rare events (defaults, operational losses).
– Exponential: time between independent events; hazard rates.
– Beta: variables bounded between 0 and 1 (e.g., probabilities, weights).
– Empirical (historical) distribution: nonparametric use of historical return samples.
Why distributions matter in investing and risk management
– Translate uncertainty into probabilities for outcomes and losses.
– Compute risk metrics (VaR, expected shortfall), scenario probabilities, stress tests.
– Inform portfolio allocation by comparing expected returns vs. probability-weighted outcomes.
– Pricing models (options, insurance) rely on assumed underlying distributions.
Practical steps to apply probability distributions (step-by-step)
1. Define the problem and variable
• Identify whether the variable is discrete or continuous (counts vs. measurements).
• Specify the time horizon and the event of interest (e.g., daily returns, defaults per year).
2. Collect and preprocess data
• Gather historical observations, clean for outliers, check for structural breaks.
• Decide on sampling frequency and length (trade-off: more data vs. nonstationarity).
3. Choose candidate distributions
• Based on variable type and domain knowledge, shortlist plausible parametric families (normal, t, lognormal, Poisson, etc.).
• Consider nonparametric (empirical) if no clear parametric fit.
4. Estimate parameters
• Use maximum likelihood estimation (MLE), method of moments, or Bayesian techniques.
• For velocity and volatility models, consider GARCH or other time-series models for heteroskedasticity.
5. Validate the fit
• Use diagnostics: QQ plots, histogram overlay, Kolmogorov–Smirnov test, Anderson–Darling, Akaike/Bayesian information criteria (AIC/BIC).
• Check tail behavior explicitly (compare empirical tails vs. model tails).
6. Model tail risk explicitly if necessary
• If data have fat tails, consider Student’s t, generalized Pareto for tail modeling, or mixture models.
• Use backtesting for VaR/ES to ensure model accuracy on worst losses.
7. Use simulation when analytical answers are difficult
• Monte Carlo simulation to generate distribution of portfolio outcomes, incorporate path-dependencies, and nonlinear instruments (e.g., options).
8. Compute risk measures and make decisions
• Compute VaR, expected shortfall, probability of loss beyond thresholds, stress-test scenarios.
• Use outputs to guide position sizing, hedging, capital buffers.
9. Update and monitor
• Re-estimate parameters regularly, especially after regime shifts.
• Monitor model performance and recalibrate when misfit is detected.
Common pitfalls and practical advice
– Don’t blindly assume normality—check empirically, especially for tails.
– Beware of sampling error: small samples produce large parameter uncertainty.
– Consider conditional heteroskedasticity (volatility clustering) common in returns—use GARCH-type models if needed.
– Use stress tests and scenario analysis to supplement probabilistic forecasts—rare but severe events may be underestimated by models.
– Overreliance on single metrics (e.g., VaR) can mislead—use multiple risk measures (VaR, expected shortfall, stress losses).
Fast fact
– The 68–95–99.7 rule: for a normal distribution, roughly 68% of observations fall within ±1σ, 95% within ±2σ, and 99.7% within ±3σ of the mean.
Putting it together: short applied example for a trader
– Goal: estimate probability of a >4% daily loss.
– Steps:
1) Collect last 1,000 daily returns.
2) Compute sample mean and standard deviation.
3) Inspect histogram and QQ plot; if tails are fatter than normal, try Student’s t.
4) Fit chosen distribution; calculate P(return < −4%).
5) Run Monte Carlo to account for stochastic volatility (e.g., GARCH) if needed.
6) Use result to size stop-losses or set margins.
The bottom line
Probability distributions are the fundamental language for quantifying uncertainty. Correctly identifying, fitting, validating, and using distributions is essential in finance—for pricing, risk measurement, and strategic decision-making. Be mindful of tail behavior, sample limitations, and changing market regimes; combine probabilistic modeling with scenario analysis and stress-testing for robust risk management.
Further reading
– Investopedia: Probability Distribution (source) —
Editor’s note: The following topics are reserved for upcoming updates and will be expanded with detailed examples and datasets.