Normal Distribution
The Gaussian bell curve: the foundation of Black-Scholes, modern portfolio theory and practically every classical statistical model in finance.
The Bell and Its Parameters
The normal distribution — also called Gaussian after Carl Friedrich Gauss, or simply "the bell" for its shape — is the most important probability distribution in statistics. It is completely characterised by just two parameters: the mean μ (mu), which determines where it is centred, and the standard deviation σ (sigma), which determines how wide it is. Its probability density formula is f(x) = (1 / σ√(2π)) × e^(−(x−μ)² / 2σ²). The bell is symmetric about μ, peaks at μ, and decays exponentially on both sides. The total area under the curve is 1 (as in any probability distribution). Its importance in finance and generally derives from the Central Limit Theorem (CLT): the sum (or average) of many independent random variables, each with an arbitrary distribution, converges to a normal distribution. That is what makes the normal ubiquitous: averages of experimental measurements, human heights, engineering measurement errors, returns of diversified portfolios — all tend toward normality. In options pricing, Black-Scholes assumes that logarithmic returns are normal (equivalent to assuming prices are lognormal, see the next concept). That assumption is mathematically elegant and analytically tractable, which allowed the 1973 model to be implemented on basic calculators. But as we will explore, reality departs from the normal in important ways.
Mean, Median and Mode
In a normal distribution, the three "measures of central tendency" — mean, median and mode — coincide exactly. The mean is the mathematical average, the median is the central value (50% of data below, 50% above), and the mode is the most frequent value. In symmetric distributions such as the normal, all three are equal. In asymmetric distributions (like the lognormal of financial prices, or return distributions with negative skew), these three measures diverge and each tells a different story. The mode is the most likely outcome in a single observation; the median is useful for distributions with large outliers (a city’s median income is more representative than the average if Jeff Bezos lives there); the mean is the centre of mass and what you would use to calculate expected values. An important property of the normal: knowing the mean and the standard deviation gives you all the information about the distribution. You need nothing else. That parsimony is mathematically beautiful but dangerous in finance — it implies assuming reality has the same elegance, which is frequently not the case. When an analyst says "returns with a mean of 8% and σ of 15%", they are implicitly assuming the distribution is normal; if it is not (and it rarely is), two distributions with those same parameters can behave very differently in the tails and therefore carry very different risk.
68-95-99.7 Revisited With Exact Mathematics
The empirical 68-95-99.7 rule is an approximation. The exact probabilities within ±Nσ of the mean are: 1σ → 68.27%, 2σ → 95.45%, 3σ → 99.73%, 4σ → 99.994%, 5σ → 99.99994%, 6σ → 99.9999998%. This exponentially rapid decline in tail probability is known as the problem of the thin tails of the normal distribution. A 5σ event should occur roughly once in 1.7 million events. In daily trading (252 days a year), that is once every 6,800 years. Yet in the actual history of financial markets, 5σ events under constant-σ models occur several times per decade. This is the main empirical criticism of models assuming normality: they systematically underestimate tail risk. Famous historical examples: Black Monday 1987 was roughly 22σ under normal assumptions (a probability of 10⁻¹⁰⁷, less likely than any physical event in the age of the universe); the March 2020 crash, with several consecutive episodes of 3 to 4σ; the 2010 flash crash. The reality is that the tails of financial distributions are fat, following distributions more like Student-t, Cauchy or Lévy stable. More sophisticated models such as Merton’s jump-diffusion (1976) or Madan’s variance gamma try to capture those tails better, at the cost of greater mathematical complexity.
Why Returns Are NOT Normal (The Empirical Reality)
Despite the supposed elegance of the normal assumption in Black-Scholes, the empirical reality of financial returns shows systematic deviations. (1) Fat tails: extreme events, both up and down, occur far more frequently than the normal predicts. The kurtosis of daily SPX returns is roughly 10-20, against the 3 a normal would have. (2) Negative skew: financial returns tend to have longer tails to the downside; crashes are more abrupt than rallies. Typical monthly SPX skew: −0.5 to −1.0 (against 0 for the normal). (3) Volatility clustering: returns are heteroscedastic; high-volatility periods cluster together (post-Lehman, COVID) and calm periods do too. Volatility is not a constant σ but a stochastic process in its own right. Models such as GARCH and stochastic volatility try to capture this. (4) Time-dependent correlations: one day’s returns are NOT entirely independent of the previous day’s; there is autocorrelation in volume and in absolute volatility (though autocorrelation in signed returns is weak in efficient markets). (5) Regime changes: mean and variance are not constant but shift between regimes (bull/bear, high-vol/low-vol). The practical consequence: models relying on normality alone underestimate maximum drawdown, underestimate tail risk, and overstate diversification benefits under stress.
Application in Options Pricing
Despite its imperfections, the normal distribution is the basis of most options pricing models. Black-Scholes assumes instantaneous logarithmic returns are normal with mean (r − σ²/2) and standard deviation σ. Equivalently, future prices are lognormal. That assumption allows the famous closed-form analytical pricing formula to be derived. The key terms d₁ and d₂ in Black-Scholes are transformations of the standard normal distribution: N(d₂) is the probability the option finishes ITM; S·N(d₁) is the expected value of the underlying conditional on exercise. The entire valuation reduces to computing areas under the normal bell. Modifications to the model to accommodate empirical reality include: (1) volatility smile/skew — instead of a constant σ, using a function σ(K, T) calibrated to market prices; (2) jump models — Merton (1976), Kou — which add a Poisson jump term on top of normal diffusion; (3) stochastic volatility models — Heston (1993), SABR, Bates — where σ itself follows a stochastic process; (4) non-parametric models based on the market’s implied density. Professional market makers use sophisticated versions of these; retail traders can work with basic Black-Scholes supplemented by intuition about skew and fat tails.
The Standard Normal Distribution and Tables
The standard normal distribution is the normal with mean 0 and standard deviation 1, N(0,1). It is denoted by the letter Z. It is convenient because any normal N(μ, σ²) can be standardised to Z via the transformation Z = (X − μ) / σ. Classical statistical tables tabulate Z-distribution probabilities to avoid computing the density integral every time. The most memorable critical values: z = 1.0 → 84.13% (area to the left), z = 1.645 → 95% (the critical value for a one-tailed test at 5%), z = 1.96 → 97.5% (critical for two tails at 5%), z = 2.326 → 99%, z = 2.58 → 99.5%, z = 3.09 → 99.9%. These values appear in: (1) confidence intervals, where the 95% interval is the mean ± 1.96σ/√n; (2) hypothesis tests, rejecting H₀ if |z| > 1.96 at 5%; (3) value at risk, where 95% VaR is roughly 1.645σ multiplied by the portfolio value. In Excel, NORM.S.DIST(z, TRUE) gives the cumulative probability; NORM.S.INV(p) returns the z value for a given percentile. In Python, scipy.stats.norm.cdf() and norm.ppf(). The standard normal is the basic building block from which practically all classical statistical tests, confidence intervals and normality tests (Shapiro-Wilk, Jarque-Bera) are constructed.