Stock Market and Probability Distributions

Why Stock Market Returns Aren’t “Normal”

A fundamental misconception, often carried over from introductory statistics, is that stock market returns follow a normal distribution, also known as the Gaussian distribution or the bell curve. This elegant, symmetric model is appealing but dangerously inaccurate for financial markets. If returns were truly normal, extreme market events like the 2008 financial crisis or a sudden 10% drop in the NIFTY 50 would be once-in-a-civilisation occurrences. In reality, we know they happen far more frequently.

The defining characteristic of financial market returns is the presence of “fat tails.” This means that extreme outcomes—both large gains and catastrophic losses—occur much more often than a normal distribution would predict. The statistical term for this is leptokurtosis. A leptokurtic distribution has a higher peak (more returns clustered around the average) and fatter tails (more returns far from the average) compared to a normal distribution.

Key Probability Distributions for Financial Modelling

While no single distribution is a perfect fit, several models are used to better capture the unique characteristics of market returns. Let’s explore the most common ones, from the simplistic to the more realistic.

1. The Normal (Gaussian) Distribution

Despite its flaws, the normal distribution is the starting point for much of financial theory. It’s defined by just two parameters: the mean (µ), representing the average return, and the standard deviation (σ), representing volatility.

• Characteristics: It is a symmetric, bell-shaped curve. In this model, roughly 68% of returns fall within one standard deviation of the mean, 95% within two, and 99.7% within three. Events beyond three standard deviations are considered exceptionally rare.
• Applicability in Trading: Its primary use is as a simplified approximation, particularly for modelling returns over very short periods or as a building block in more complex models. However, relying on it for risk management is a critical error. It systematically underestimates the probability of large drawdowns.

2. Student’s t-Distribution

The Student’s t-distribution is a step up in realism. It resembles the normal distribution but features heavier tails, making it better suited for capturing the frequent extreme events seen in financial data.

• Characteristics: The key parameter here is (nu), the degrees of freedom. A lower value of results in fatter tails. As approaches infinity, the t-distribution converges to the normal distribution. This flexibility allows it to be “tuned” to the observed data.
• Applicability in Trading: It provides a more realistic model for stock returns than the normal distribution. When backtesting strategies, using a t-distribution to model the return series can give a more sober assessment of potential risks and drawdowns.

3. Log-Normal Distribution

This distribution is used to model asset prices themselves, rather than their returns. It assumes that the logarithm of the asset’s price is normally distributed. This is a critical distinction.

• Characteristics: The log-normal distribution is skewed to the right and is only defined for positive values. This makes intuitive sense for stock prices, as a stock’s price cannot go below zero. It also naturally incorporates the effect of compounding returns over time.
• Applicability in Trading: It is the foundation of the Black-Scholes options pricing model. It helps in understanding that a stock going from Rs. 100 to Rs. 200 (a 100% gain) is a multiplicative process, not an additive one.

4. Cauchy Distribution

The Cauchy distribution is an extreme case of a fat-tailed distribution. It’s a theoretical model that is rarely used in practice for direct financial modelling but serves as a cautionary tale.

• Characteristics: Its tails are so fat that its mean and variance are mathematically undefined. This implies that in a Cauchy world, extreme fluctuations are so common that calculating an “average” return or “standard” volatility is meaningless.
• Applicability in Trading: While not used for direct modelling, it’s a useful mental model for “black swan” events or market regimes where historical averages break down completely. It reminds us that under certain conditions, the market can behave in ways that seem to have no precedent.

A More Dynamic Approach: GARCH Models

The distributions discussed so far are static; they assume the parameters (like volatility) are constant over time. This is another major flaw. Anyone watching the markets knows that volatility is not constant. There are quiet periods of low volatility and frantic periods of high volatility.

This phenomenon is called volatility clustering. High volatility tends to be followed by more high volatility, and low volatility by more low volatility.

GARCH (Generalized Autoregressive Conditional Heteroskedasticity) models are designed specifically to capture this. Instead of assuming a fixed volatility (), a GARCH model forecasts volatility for the next period based on past information.

A standard GARCH(1,1) model defines the variance for the next period () as a function of three components:

• (omega): The long-run average variance.
• (alpha term): The “shock” from the previous period. A large price move yesterday increases today’s expected volatility.
• (beta term): The forecasted variance from the previous period. This represents the “persistence” of volatility.

GARCH doesn’t replace the return distributions mentioned earlier. Rather, it provides a dynamic forecast for their volatility parameter, making the overall model far more adaptive and realistic.

Conclusion: What This Means for the Practical Trader

Understanding these concepts is not a mere academic exercise. It has profound implications for survival and success in trading:

1. Respect Tail Risk: The market can and will make moves that seem statistically impossible. Your risk management—specifically your stop-loss placement and position sizing—is your only real defence against a fat-tail event wiping out your account.
2. Question Assumptions: Be highly sceptical of any system, software, or guru who implicitly assumes normal returns. Their risk calculations are likely flawed and overly optimistic.
3. Adapt to Volatility Regimes: Volatility is not static. A strategy that works well in a low-volatility environment may get destroyed in a high-volatility one. Recognising the current volatility regime (e.g., using indicators like ATR) is a critical skill.

Ultimately, the market’s statistical nature is complex and ever-changing. While we can create increasingly sophisticated models, a trader must accept that there will always be an element of uncertainty. The goal is not to predict the future perfectly but to build a robust trading strategy that can survive—and even thrive—in a world of fat tails and clustered volatility.