1. Does the MGF Exist?
A fundamental distinction in probability theory and econometrics is whether a distribution possesses a moment generating function (MGF).
Let
\[ M_X(t) = E[e^{tX}] \]
denote the MGF of a random variable \(X\).
A distribution \(F\) is called heavy-tailed if
\[ M_X(t) = \infty, \qquad \forall t > 0 \]
In contrast, if there exists some \(t>0\) such that
\[ M_X(t) < \infty, \]
the distribution is said to be light-tailed.
This criterion acts as a natural “death line”: once the MGF ceases to exist, classical exponential large-deviation tools break down, and extreme outcomes become fundamentally more likely.
2. The Hierarchy of Heavy Tails
Heavy-tailed distributions themselves form a structured hierarchy.
A commonly used nesting is
\[ \text{Regularly Varying} \subset \text{Subexponential} \subset \text{Long-tailed} \subset \text{Heavy-tailed}. \]
Each step corresponds to progressively weaker assumptions about how slowly the tail decays.
2.1 Long-tailed distributions
A distribution belongs to the long-tailed class if
\[ \frac{\bar F(x+y)}{\bar F(x)} \to 1 \quad \text{as } x \to \infty \]
for every fixed \(y\), where
\[ \bar F(x) = P(X>x) \]
is the survival function.
Intuitively, adding a fixed constant \(y\) barely changes the tail probability when \(x\) is large.
2.2 Subexponential distributions and the “Single Big Jump”
A central subclass of heavy tails is the subexponential class, defined by
\[ \overline{F^{*2}}(x) \sim 2\bar F(x), \quad x \to \infty. \]
Equivalently,
\[ P(X_1 + X_2 > x) \sim 2P(X > x) \]
for independent copies \(X_1, X_2\).
This property leads to the Single Big Jump Principle:
When a sum of subexponential variables becomes extremely large, it is typically caused by one extreme observation, rather than by the accumulation of many moderate values.
2.3 Regular variation (power-law tails)
The most widely studied heavy tails are regularly varying tails, characterized by
\[ \frac{\bar F(tx)}{\bar F(x)} \to t^{-\alpha}, \quad t>0. \]
Equivalently,
\[ \bar F(x) = x^{-\alpha} L(x), \]
where \(L(x)\) is a slowly varying function.
Regular variation provides a tractable mathematical structure and plays a central role in extreme value theory and financial tail modeling.
3. Light Tails: The “Conspiracy of Smalls”
Light-tailed distributions behave very differently from heavy-tailed ones.
For light tails, the probability of producing one extremely large observation is exponentially small. As a result, extreme sums typically arise through the combined effect of multiple moderately large values, rather than a single outlier.
For example, suppose
\[ f(x) \propto e^{-x^2}. \]
To produce a sum of \(10\):
- Single jump: \((10 + 0)\) gives probability \(\propto e^{-100}\)
- Two moderate values: \((5 + 5)\) gives probability \(\propto e^{-50}\)
Since \(e^{-50} \gg e^{-100}\), the second scenario dominates.
This phenomenon explains why Gaussian-type distributions do not satisfy the subexponential property.
4. Summary of Tail Classes
| Class | Mathematical Criterion | Typical Examples |
|---|---|---|
| Light-tailed | \(M_X(t)<\infty\) for some \(t>0\) | Normal, Exponential |
| Heavy-tailed | \(M_X(t)=\infty\) for all \(t>0\) | Lognormal, Pareto |
| Long-tailed | \(\bar F(x+y)/\bar F(x)\to1\) | Pareto, Lognormal |
| Subexponential | \(\overline{F^{*2}}(x)/\bar F(x)\to2\) | Pareto, Lognormal |
| Regularly varying | \(\bar F(tx)/\bar F(x)\to t^{-\alpha}\) | Pareto, Student-t |
5. Conclusion
Tail behavior determines how extreme events arise.
- In light-tailed environments, extreme outcomes result from a conspiracy of many moderate deviations.
- In subexponential environments, extremes are dominated by a single big jump.
Understanding which regime a model belongs to is crucial for applications ranging from risk management to econometric modeling of financial volatility.