A normal distribution (Gaussian distribution or “bell curve”) is a continuous probability distribution in which values symmetrically cluster around a central mean. The shape is unimodal and symmetric: the mean, median and mode are equal, and the spread of the distribution is governed entirely by the standard deviation. Many statistical methods and financial models assume normality because it is mathematically tractable and because, by the Central Limit Theorem, averages of many independent, identically distributed random variables tend toward a normal distribution.
Key takeaways
– Shape: symmetric, bell-shaped curve centered at the mean (μ).
– Parameters: mean (μ) and standard deviation (σ).
– Empirical rule: ~68.3% within ±1σ, ~95.4% within ±2σ, ~99.7% within ±3σ.
– Skewness = 0 and kurtosis = 3 for a true normal distribution.
– Widely used in finance for modeling returns, risk measures and hypothesis testing—but has important limitations (fat tails, skew) for many financial series.
– Alternatives and robustness checks (t-distribution, EVT, bootstrap, GARCH) are often needed in practice.
What is meant by the normal distribution?
– Definition: A continuous distribution where the probability density function (PDF) is symmetric about the mean and declines smoothly as you move away from the mean. Observations close to the mean are most likely; extreme values become exponentially unlikely.
– Z-score: A standardized normal variable Z = (X − μ) / σ converts any normal distribution to the standard normal (mean 0, SD 1), enabling use of standard normal tables.
The formula for the normal distribution
– Probability density function (PDF)
f(x) = (1 / (σ √(2π))) * exp( −(x − μ)² / (2σ²) )
Where μ is the mean and σ is the standard deviation.
– Interpretation: The PDF gives relative likelihood; probabilities over intervals come from integrating f(x) over that interval. In practice we use z-scores and standard normal tables or software to compute probabilities.
Understanding the empirical rule in normal distribution
– 68.27% (approx) of observations lie within ±1σ of the mean.
– 95.45% lie within ±2σ.
– 99.73% lie within ±3σ.
– Use case: Quick, back-of-the-envelope checks for whether an observation is “rare” (e.g., a 3-sigma event should be very rare under normality, ≈0.3% of outcomes).
Analyzing skewness in normal distribution
– Skewness measures asymmetry. For the normal distribution, skewness = 0.
– Positive skew: long right tail (more extreme large values). Negative skew: long left tail.
– Financial returns often show nonzero skewness (right or left), violating strict normality.
Exploring kurtosis in normal distribution
– Kurtosis measures tail weight/peakiness. For the normal distribution, kurtosis = 3 (often reported as excess kurtosis = 0).
– Excess kurtosis > 0 (leptokurtic) indicates “fat tails” — extreme outcomes are more frequent than under normality. This is common in asset returns and is a source of tail risk.
– Excess kurtosis 3, so extreme losses/gains happen more often than normality predicts.
– Skewness: Distributions of returns can be asymmetric (negative skew common in equity drops), invalidating the zero-skew assumption.
– Non-stationarity: Return distributions change over time (volatility clustering, regime shifts), so fixed μ and σ estimates can be misleading.
– Serial correlation and conditional heteroskedasticity: Returns often exhibit volatility clustering (GARCH effects), contradicting i.i.d. assumptions.
– Jumps and discontinuities: Market shocks (crashes, news) produce jumps that normal models understate.
– Misleading risk estimates: VaR and other risk metrics based strictly on normality can understate tail risk, producing overconfident capital or position-sizing decisions.
Alternatives and complements to normality (when normal is inadequate)
– Student’s t-distribution: heavier tails; controlled by degrees-of-freedom parameter.
– Log-normal for asset prices (if positive and multiplicative growth).
– Mixture distributions or regime-switching models to capture multiple market states.
– Generalized (hyperbolic) or stable distributions for extreme tails.
– Extreme Value Theory (EVT): focus on tail behavior to estimate probabilities of extreme events.
– GARCH-family models: capture time-varying volatility; can be combined with heavy-tailed error distributions.
– Non-parametric/historical simulation and bootstrap: rely less on distributional assumptions.
Practical example — quick VaR approximation using normality
– Suppose daily return mean μ̂ = 0.05% and σ̂ = 1%. A one-day 99% VaR (left tail) under normality: find z0.01 ≈ −2.33, so VaR ≈ μ̂ + z0.01·σ̂ ≈ 0.05% + (−2.33)(1%) ≈ −2.28%. Interpretation: there is an estimated 1% chance of a daily loss worse than about 2.28%. Caveat: if returns have fat tails, actual 1% losses may be larger and more frequent.
The bottom line
The normal distribution is central to statistics and many financial tools because of its simplicity and the central limit theorem. It provides an intuitive framework (mean and standard deviation) and practical rules (empirical rule, z-scores). However, many financial time series deviate from normality—exhibiting skewness, kurtosis, volatility clustering and jumps—so analysts must test normality, use normal-based methods cautiously, and apply heavy-tailed models or non-parametric methods where tails and changes in volatility matter. Always validate assumptions through visual checks, statistical tests and backtesting before relying on normal-based risk estimates or trading rules.
Sources and further reading
– Investopedia, “Normal Distribution,” Investopedia.com. (Article background and financial discussion.)
– DePaul University, “Normal Distribution: Origin of the Name.” (Historical note.)
– Run a short worked example using your own return series (compute μ, σ, z-scores, histograms and normality tests), or
– Show R/Python code snippets to compute probabilities, VaR, and apply normality tests and alternatives (t-distribution, EVT, bootstrap). Which would you prefer?