Probability Distributions in Finance: Normal, Lognormal, Binomial, and Student’s t

Published: May 9, 2026
Ryan O'Connell, CFA, FRM
Article by Ryan O'Connell, CFA, FRM

Probability distributions in finance provide the vocabulary for describing uncertainty. Whether you’re estimating the likelihood of a stock losing more than 10% in a year, modeling how asset prices evolve over time, or counting defaults in a loan portfolio, the distribution you choose determines the assumptions built into your analysis — and the risks you might be missing. This guide covers the four distributions most commonly used in finance: normal, lognormal, binomial, and Student’s t.

What Are Probability Distributions in Finance?

A probability distribution describes how the possible values of a random variable are distributed. In finance, random variables include asset returns, stock prices, default counts, and sample statistics. Each distribution is defined by parameters that control its shape, center, and spread.

Key Concept

A probability distribution tells you how likely each possible outcome is. For continuous variables (like returns), you work with a probability density function (PDF). For discrete variables (like the number of defaults), you work with a probability mass function (PMF).

The cumulative distribution function (CDF) gives the probability that the variable is less than or equal to a given value. For continuous distributions, the CDF is the integral of the PDF; for discrete distributions, it’s the cumulative sum of the PMF.

Probability Density Function (PDF)

  • Used for continuous variables
  • Height represents relative likelihood
  • Area under the curve equals probability
  • Example: stock returns, interest rates

Probability Mass Function (PMF)

  • Used for discrete variables
  • Height represents exact probability
  • Sum of all masses equals 1
  • Example: default counts, up-moves

Normal Distribution

The normal distribution — also called the Gaussian or bell curve — is the most widely used distribution in finance. It’s symmetric around its mean, with tails that extend to infinity in both directions. Many risk models, including Value at Risk (VaR) and portfolio optimization, assume returns are normally distributed.

The Normal PDF

Normal Probability Density Function
f(x) = (1 / (σ√(2π))) × e−(x−μ)2/(2σ2)
Where μ is the mean and σ is the standard deviation

The normal distribution is fully characterized by two parameters: the mean (μ) determines where the distribution is centered, and the standard deviation (σ) determines how spread out it is.

Z-Scores and the 68-95-99.7 Rule

A z-score standardizes any normal variable to the standard normal distribution (mean = 0, standard deviation = 1), making it easy to look up probabilities in a table or use a calculator.

Z-Score Formula
z = (x − μ) / σ
Number of standard deviations from the mean

The empirical rule states that approximately 68% of observations fall within 1 standard deviation of the mean, 95% within 2, and 99.7% within 3. This rule provides quick intuition but understates the probability of extreme events in real financial data.

Pro Tip

Z-tables give you P(Z ≤ z) — the probability of observing a value less than or equal to your z-score. For P(Z > z), subtract from 1. For two-tailed probabilities, double the one-tail value.

Why Returns Are Often Modeled as Normal

Short-horizon log returns tend to be approximately normal due to the Central Limit Theorem: when many small, independent factors combine additively, the result approaches a normal distribution. This is why parametric VaR calculations and mean-variance portfolio theory assume normality.

However, the normal distribution has thin tails — it systematically underestimates the probability of extreme moves. Real financial returns exhibit fat tails: crashes and spikes happen more often than the normal model predicts. For simulation and derivative pricing, see the Monte Carlo simulation methodology, which builds on these distributional foundations.

Watch: Probability Distributions in Action — Monte Carlo Simulation

Lognormal Distribution

While returns may be approximately normal, asset prices cannot be. Prices are strictly positive — a stock can’t fall below zero — but the normal distribution assigns probability to negative values. The lognormal distribution solves this problem.

The Log-Return / Price Relationship

If continuously compounded log-returns are normally distributed, then the price level follows a lognormal distribution. This is the foundation of geometric Brownian motion, which underlies the Black-Scholes model and many other pricing frameworks.

Lognormal Relationship
If Y ~ N(μ, σ2), then ST = S0 × eY is lognormal
Y is the continuously compounded log-return; μ and σ are the mean and standard deviation of the log-return

A critical implication: the expected value of a lognormal variable is higher than its median.

Expected Value of a Lognormal Variable
E[ST] = S0 × eμ + σ2/2
The σ²/2 term accounts for the convexity of the exponential function
Compounding Bias

Because the lognormal distribution is right-skewed, the arithmetic mean of prices exceeds the median. Higher volatility widens this gap. Due to the σ²/2 term in E[ST], the expected arithmetic return exceeds the expected log-return — and the gap grows with volatility. This is why average compound returns underperform average arithmetic returns over time, especially for volatile assets. Confusing the expected log-return (μ) with the expected arithmetic return is a common source of error.

Binomial Distribution

The binomial distribution describes the number of successes in a fixed number of independent trials, where each trial has two possible outcomes (success or failure) with constant probability. In finance, it’s used to model counts of discrete events: how many loans default, how many trades are profitable, or how many periods an asset moves up.

Bernoulli Trials and the PMF

A Bernoulli trial is a single experiment with two outcomes: success (probability p) or failure (probability 1 − p). The binomial distribution counts the number of successes across n independent Bernoulli trials.

Binomial Probability Mass Function
P(X = k) = C(n,k) × pk × (1−p)n−k
C(n,k) = n! / (k!(n−k)!) is the binomial coefficient
Binomial Mean and Variance
E[X] = np     Var(X) = np(1−p)
The expected number of successes and the variance around it

Applications include estimating the expected number of defaults in a homogeneous loan portfolio or modeling the probability of achieving a certain hit rate across a sequence of trades. For option pricing tree models that use a different probabilistic framework, see binomial option pricing.

Independence Assumption

The binomial model assumes each trial is independent. In credit portfolios, defaults tend to cluster during recessions — violating independence. Using the binomial directly understates the probability of multiple simultaneous defaults. Copula models and other correlation-adjusted approaches address this.

Student’s t Distribution

The Student’s t distribution appears in small-sample inference when the population standard deviation is unknown. It looks similar to the normal distribution but has fatter tails, reflecting the additional uncertainty from estimating σ with limited data.

When to Use t Instead of Normal

When you’re constructing confidence intervals or conducting hypothesis tests on a sample mean with unknown population variance, the appropriate distribution is t with n − 1 degrees of freedom — not the standard normal. As degrees of freedom increase (roughly past 30), the t distribution converges to normal.

t-Statistic
t = (x̄ − μ0) / (s / √n)
Where x̄ is the sample mean, s is the sample standard deviation, and df = n − 1

The fatter tails of the t distribution mean that extreme values are more probable than under normal assumptions — which makes confidence intervals wider and hypothesis tests more conservative when sample sizes are small. For detailed mechanics on confidence intervals and hypothesis testing, see confidence intervals and hypothesis testing.

The t distribution also appears in fat-tailed return modeling. Some practitioners use t-distributed returns (with low degrees of freedom) as a more realistic alternative to normality. For deeper coverage of skewness and kurtosis in returns, see skewness and kurtosis.

Worked Examples

Example 1: Normal — S&P 500 Loss Probability

Setup: Assume the S&P 500’s annualized return is normally distributed with μ = 9% and σ = 16%. What is the probability of losing more than 10% in a year?

Step 1: Calculate the z-score for a −10% return:

z = (−10% − 9%) / 16% = −19% / 16% = −1.19

Step 2: Look up P(Z < −1.19) in a z-table:

P(Z < −1.19) ≈ 0.117 (about 11.7%)

Interpretation: Under the normal assumption, there’s roughly a 1-in-9 chance of an annual loss worse than 10%. The actual historical frequency is higher — real returns have fatter tails than the normal model predicts.

Example 2: Lognormal — Microsoft Stock Price Distribution

Setup: Microsoft (MSFT) trades at $400 today. Assume the continuously compounded annual log-return Y is normally distributed with μ = 8% and σ = 25%.

Question: What is the median price in one year? What is the expected price?

Median price: The median of a lognormal variable occurs at S0eμ (since 50% of log-returns are below μ).

Median ST = $400 × e0.08 = $400 × 1.0833 = $433.31

Expected price:

E[ST] = $400 × e0.08 + (0.25)2/2 = $400 × e0.08 + 0.03125 = $400 × e0.11125 = $400 × 1.1177 = $447.08

Key insight: The expected price ($447.08) exceeds the median ($433.31) because the lognormal distribution is right-skewed. Volatility creates a gap between typical and average outcomes.

Example 3: Binomial — Defaults in a High-Yield Bond Portfolio

Setup: A portfolio contains n = 20 single-B rated corporate bonds (similar credit quality to issuers like AMC Entertainment or Carvana), each with a one-year default probability of p = 0.05 (5%). Assume defaults are independent.

Question: What is the probability of exactly 0 defaults? Exactly 2 defaults? 3 or more defaults?

P(X = 0): No loans default.

P(X = 0) = (0.95)20 = 0.358 (35.8%)

P(X = 1):

P(X = 1) = C(20,1) × (0.05)1 × (0.95)19 = 20 × 0.05 × 0.377 = 0.377 (37.7%)

P(X = 2):

P(X = 2) = C(20,2) × (0.05)2 × (0.95)18 = 190 × 0.0025 × 0.397 = 0.189 (18.9%)

P(X ≥ 3):

P(X ≥ 3) = 1 − P(0) − P(1) − P(2) = 1 − 0.358 − 0.377 − 0.189 = 0.076 (7.6%)

Expected defaults: E[X] = np = 20 × 0.05 = 1.0 default

Note: The independence assumption is critical. In a recession, defaults cluster — the binomial model would understate tail risk.

Choosing the Right Distribution

Each distribution fits a specific type of financial variable. Choosing incorrectly embeds hidden assumptions into your model.

Normal

  • Use for: Short-horizon log returns, regression residuals
  • Parameters: μ (mean), σ (std dev)
  • Watch for: Underestimates tail risk; allows negative values

Lognormal

  • Use for: Asset prices, terminal wealth
  • Parameters: μ, σ of log-returns
  • Watch for: Right-skew; mean > median; still misses extreme tails

Binomial

  • Use for: Counts of successes/failures over n trials
  • Parameters: n (trials), p (probability)
  • Watch for: Requires independence; not for continuous outcomes

Student’s t

  • Use for: Small-sample inference; fat-tailed return modeling
  • Parameters: df (degrees of freedom)
  • Watch for: Converges to normal as df grows

Decision rule: If the variable is continuous and can go negative → start with normal. If continuous but bounded below by zero → lognormal. If you’re counting discrete events → binomial. If you’re doing inference on a mean with limited data → t.

How to Apply Probability Distributions in Practice

Follow this four-step workflow when building distribution-based models:

  1. Identify the variable type. Is it a return, a price, a count, or a sample statistic?
  2. Select a candidate distribution. Use the decision rule above as a starting point.
  3. Validate against data. Plot histograms, compute summary statistics (mean, variance, skewness, kurtosis), and run goodness-of-fit tests.
  4. Stress-test assumptions. What happens if tails are fatter than assumed? If parameters shift? If independence breaks down?
Try the Normal Distribution Calculator

For binomial probability calculations, see the Binomial Distribution Calculator. For a deeper dive into portfolio analytics and risk modeling, explore the Portfolio Analytics & Risk Management course.

Common Mistakes

Mistake #1: Modeling Prices as Normal Instead of Lognormal

Stock prices can’t go below zero, but the normal distribution assigns probability to negative values. Always use lognormal for prices; reserve normal for returns. This mistake can lead to nonsensical model outputs, especially at long horizons or high volatilities.

Mistake #2: Using a z-table when the sample is small. With n ≤ 30 and an unknown population standard deviation, the correct sampling distribution is t with n − 1 degrees of freedom, not the standard normal. Using z understates uncertainty and produces artificially narrow confidence intervals.

Mistake #3: Assuming independence in binomial default models. The textbook binomial assumes each trial is independent. Real loan portfolios experience correlated defaults during economic downturns. Using the binomial directly understates joint-default risk — the very scenario that matters most.

Mistake #4: Calibrating distributions on a calm sample. Estimating μ and σ from 2017–2019 data and applying them to a 2020 stress scenario is a recipe for model failure. Distributions describe a regime, not a permanent law. Parameters are conditional on the sample period — and the future may look nothing like the past.

Limitations of Probability Distributions in Finance

Model Risk

Every distribution is a simplification. Choosing a distribution means choosing which features of reality to capture and which to ignore. The risk that your model’s assumptions are wrong — and that you won’t know until it’s too late — is called model risk.

Tail underestimation. The normal distribution underestimates the probability of extreme events. Real returns have fatter tails — crashes and spikes occur more frequently than predicted. For tail-aware modeling, consider the t distribution, mixture models, or extreme value theory.

Regime changes. Markets move through different regimes — calm periods and crises — with different return characteristics. A single distribution calibrated on a long sample may not describe any regime well.

Time-varying volatility. Volatility clusters: large moves tend to follow large moves. Standard i.i.d. distributional assumptions miss this autocorrelation in squared returns.

Serial dependence. Returns may exhibit momentum or mean reversion over various horizons. Distributions that assume independence across time can miss these patterns.

Parameter instability. Mean and variance themselves change over time. The μ and σ you estimate from historical data may not hold going forward.

Distributions are tools, not truths. They provide a framework for thinking about uncertainty — but the assumptions embedded in each choice have real consequences. For deeper coverage of fat tails and asymmetric returns, see skewness and kurtosis.

Frequently Asked Questions

A probability density function (PDF) describes the relative likelihood of a continuous random variable taking on a particular value. The height of the PDF indicates relative probability density, but you need to integrate over a range to get an actual probability. A cumulative distribution function (CDF) gives the probability that the variable is less than or equal to a given value — it’s the running integral of the PDF. For discrete variables, the analogous functions are the probability mass function (PMF) and its cumulative sum.

Stock prices cannot fall below zero, but the normal distribution places probability on negative values. The lognormal distribution is bounded below by zero, making it a natural fit for prices. Mathematically, if continuously compounded log-returns are normally distributed, prices follow a lognormal distribution. This relationship underlies geometric Brownian motion and the Black-Scholes option pricing model. However, even the lognormal distribution underestimates extreme tail events.

Use the t distribution when constructing confidence intervals or conducting hypothesis tests on a sample mean where the population standard deviation is unknown and the sample size is small (typically n ≤ 30). The t distribution has fatter tails than the normal, reflecting the additional uncertainty from estimating σ with limited data. As degrees of freedom increase, the t distribution converges to the normal. Some practitioners also use t-distributed returns to model fat tails in asset prices.

No — the normal distribution systematically underestimates the probability of extreme events. Financial returns exhibit excess kurtosis: large positive and negative moves occur more frequently than the normal model predicts. For tail-aware modeling, practitioners use the Student’s t distribution (with low degrees of freedom), mixture models, or extreme value theory. However, even these approaches require assumptions that may fail during unprecedented market stress.

The binomial distribution models the count of successes across a fixed number of independent trials with constant success probability. In finance, applications include estimating the expected number of defaults in a homogeneous loan portfolio, calculating the probability of a trading strategy achieving a certain win rate, or modeling the number of up-moves over a fixed horizon. The key assumption is independence across trials — when events are correlated (as defaults often are), the binomial model understates risk.

Short-horizon log returns are often approximately normal in the sense that the mean and variance describe most of the distribution adequately, and the Central Limit Theorem explains why aggregated returns tend toward normality. However, “approximately” is doing heavy lifting: the tails are heavier than normal (excess kurtosis), returns may be slightly skewed, and volatility clusters over time. The normal approximation works well for everyday risk estimation but fails precisely when it matters most — during market crises.

Disclaimer

This article is for educational and informational purposes only and does not constitute investment advice. Probability distributions are theoretical models with simplifying assumptions; actual market behavior may differ significantly. Always conduct your own research and consult a qualified financial advisor before making investment decisions.