Corrections to the Central Limit Theorem for Heavy-Tailed Probability Densities

Henry Lam, José Blanchet, Damian Burch, Martin Z. Bazant

Abstract

Classical Edgeworth expansions provide asymptotic correction terms to the Central Limit Theorem (CLT) up to an order that depends on the number of moments available. In this paper, we provide subsequent correction terms beyond those given by a standard Edgeworth expansion in the general case of regularly varying distributions with diverging moments (beyond the second). The subsequent terms can be expressed in a simple closed form in terms of certain special functions (Dawson's integral and parabolic cylinder functions), and there are qualitative differences depending on whether the number of moments available is even, odd or not an integer, and whether the distributions are symmetric or not. If the increments have an even number of moments, then additional logarithmic corrections must also be incorporated in the expansion parameter. An interesting feature of our correction terms for the CLT is that they become dominant outside the central region and blend naturally with known large-deviation asymptotics when these are applied formally to the spatial scales of the CLT.

Introduction

The Central Limit Theorem (CLT) is the bedrock of statistics, establishing that the sum of many independent and identically distributed (i.i.d.) variables tends toward a normal (Gaussian) distribution. However, the CLT is a first-order approximation. To achieve higher accuracy, mathematicians traditionally use Edgeworth expansions, which add correction terms to account for skewness and kurtosis.

A fundamental limitation of standard Edgeworth expansions is their reliance on finite higher-order moments (such as the 3rd or 4th moment). In the context of "heavy-tailed" or "fat-tailed" distributions—frequently encountered in finance, physics, and insurance—these moments often diverge. When moments become infinite, the standard expansion "hits a wall" and fails to provide further refinements. This paper addresses this gap by deriving new correction terms specifically for distributions with power-law tails (regularly varying distributions) where higher moments do not exist.

Methodology

The research employs a rigorous Fourier-based approach to analyze the characteristic function (the Fourier transform of the probability density). The core innovation lies in the decomposition of the characteristic function into two distinct parts:

  1. Analytic Component: This leads to the standard Edgeworth terms, typically expressed via Hermite polynomials.
  2. Non-Analytic Component: This arises directly from the heavy tail of the distribution.

By performing an inverse Fourier transform on the non-analytic component, the authors derive corrections that are expressed through special functions: Dawson’s integral and parabolic cylinder functions. This derivation encompasses various scenarios, including symmetric tails, asymmetric tails where one side is "heavier," and cases where the power-law exponent is an integer versus a non-integer.

Key Theoretical Insights

The Parity Effect and Logarithmic Corrections

The nature of the correction terms changes qualitatively depending on the tail exponent $\alpha$. The mathematical forms differ significantly depending on whether $\alpha$ is even, odd, or a non-integer. A particularly subtle finding is that if a distribution has an even number of finite moments, the expansion requires additional logarithmic terms ($\log n$) to maintain accuracy.

The "Smooth Transition" to Large Deviations

Traditionally, the CLT describes the "center" of a distribution, while Large Deviation Theory describes the "far tails." This paper demonstrates that the new correction terms naturally "blend" into large-deviation asymptotics. This provides a unified mathematical bridge between the two regimes. While standard Edgeworth terms eventually vanish as one moves further from the mean, these new heavy-tail corrections become the dominant source of error in the "moderate deviation" region.

Key Results

The authors formalize their findings in Theorems 1 and 2, providing explicit, closed-form formulas for the probability density of a normalized sum ($S_n/n^{1/2}$). These formulas include the standard Gaussian term, the existing Edgeworth terms (up to the last finite moment), and the new heavy-tail correction term $F(x, n)$.

Validation was performed through mathematical proofs and comparisons to known Pareto distributions. The researchers proved that for certain classes of distributions, such as the symmetric Pareto, their refined expansion is valid uniformly over a wide range of spatial scales, specifically for $x \leq \sqrt{(\alpha - 2) \log n}$. Furthermore, they demonstrated that their non-analytic correction term matches the large-deviation results of Rozovskii (1989) exactly when applied to the spatial scales of the CLT.

Limitations and Future Work

The current framework is specifically tailored for distributions with "regularly varying" tails (power laws). Consequently, it may not apply to distributions with different tail behaviors, such as log-normal or "super-heavy" tails. Additionally, the proofs assume the underlying densities are continuous and satisfy specific differentiability conditions on their slowly varying components.

Future research may extend this framework to stable limits—where the sum converges to a non-Gaussian Levy distribution—and to more complex stochastic systems, such as queueing models in heavy traffic environments.

Summary updated

Abstract

Classical Edgeworth expansions provide asymptotic correction terms to the Central Limit Theorem (CLT) up to an order that depends on the number of moments available. In this paper, we provide subsequent correction terms beyond those given by a standard Edgeworth expansion in the general case of regularly varying distributions with diverging moments (beyond the second). The subsequent terms can be expressed in a simple closed form in terms of certain special functions (Dawson's integral and parabolic cylinder functions), and there are qualitative differences depending on whether the number of moments available is even, odd or not an integer, and whether the distributions are symmetric or not. If the increments have an even number of moments, then additional logarithmic corrections must also be incorporated in the expansion parameter. An interesting feature of our correction terms for the CLT is that they become dominant outside the central region and blend naturally with known large-deviation asymptotics when these are applied formally to the spatial scales of the CLT.
BibTeX
@article{lam2011corrections,
  author = {Henry Lam and José Blanchet and Damian Burch and Martin Z. Bazant},
  title = {{Corrections to the Central Limit Theorem for Heavy-Tailed Probability Densities}},
  journal = {Journal of Theoretical Probability},
  year = {2011},
  doi = {10.1007/s10959-011-0379-y},
}
Figure