An almost sure central limit theorem of products of partial sums for ρ--mixing sequences

Xili Tan, Ying Zhang, Yong ZhangJournal of Inequalities and Applications 2012 · DOI

Introduction

In probability theory, the Central Limit Theorem (CLT) is a cornerstone, describing how the sum of independent and identically distributed (i.i.d.) random variables converges to a normal distribution. The Almost Sure Central Limit Theorem (ASCLT) provides a stronger form of convergence, focusing on the logarithmic average of these occurrences over time. While well-understood for independent variables, many real-world stochastic processes exhibit dependence. This paper extends the ASCLT framework to $\rho^-$-mixing sequences, a broad class of dependent variables that includes negatively associated sequences.

Problem Statement: From Sums to Products

Most limit theorems focus on the behavior of sums. However, products of partial sums present a more complex challenge. The authors address whether a predictable logarithmic pattern emerges when multiplying the cumulative totals of a dependent sequence. Specifically, they investigate a strictly stationary $\rho^-$-mixing sequence $\{X_n, n \geq 1\}$ of positive random variables.

The $\rho^-$-mixing condition is particularly significant because it encompasses several other types of dependence, such as Negatively Associated (NA) and $\rho^*$-mixing sequences. This makes the findings applicable to a wide range of stochastic models where variables are not independent but exhibit specific decay in their correlation.

Methodology

To bridge the gap between simple sums and complex products, the authors utilize several advanced tools from probability theory:

  1. Logarithmic Transformation: By taking the logarithm of the product of partial sums, the problem is converted into a sum of logarithms. This allows the application of techniques developed for weighted sums.
  2. Moment Inequalities: The proof relies on Rosenthal-type moment inequalities specifically adapted for $\rho^-$-mixing sequences. These are used to bound the variance of the terms and ensure stability in the limit.
  3. Weighted Sum Convergence: Before proving the ASCLT, the authors establish a standard CLT for weighted sums of these dependent variables (Theorem 1.1).
  4. Borel-Cantelli Lemma: To achieve "almost sure" (a.s.) convergence—meaning the limit holds with probability 1—the authors apply the Borel-Cantelli lemma to the sequence of indicators.

The primary innovation lies in the delicate estimation required to handle the $\rho^-$-mixing dependence structure alongside the functional form of products, which is more sensitive to fluctuations than standard additive sums.

Key Results

The central contribution of the paper is Theorem 1.2. For a strictly stationary $\rho^-$-mixing sequence $\{X_n\}$ of positive random variables with mean $\mu$ and coefficient of variation $\gamma$, the following holds:

$$\lim_{n \to \infty} \frac{1}{\log n} \sum_{k=1}^{n} \frac{1}{k} I \left\{ \left( \prod_{i=1}^{k} \frac{S_i}{i\mu} \right)^{\frac{1}{\gamma \sigma_k}} \leq x \right\} = F(x) \text{ a.s.}$$

In this expression:

  • $S_i$ represents the partial sum $\sum_{j=1}^i X_j$.
  • $\sigma_k^2$ is a variance-related scaling factor.
  • $F(x)$ is the distribution function of $e^{\sqrt{2}\mathcal{N}}$, where $\mathcal{N}$ is a standard normal variable.

This result confirms that the product of partial sums for these dependent sequences follows a log-normal-like distribution in the limit, provided the mixing rate decays sufficiently fast (specifically $O(\log^{-\delta} n)$ for $\delta > 1$).

Discussion and Limitations

While this work significantly broadens the scope of the ASCLT, several avenues for future research remain:

  • Stationarity: The current proof assumes strict stationarity. Extending these results to non-stationary sequences, where statistical properties evolve over time, is a logical next step.
  • Moment Conditions: The theorem requires the existence of moments $E|X_1|^r < \infty$ for $r > 2$. The behavior of "heavy-tailed" distributions, where such moments may be infinite, remains an open question.
  • Mixing Rates: The requirement for a specific decay rate of the mixing coefficient suggests that very "sticky" or long-memory dependence might break the theorem.
  • Applications: The theoretical framework established here has potential applications in econometrics and signal processing, where data often exhibit the $\rho^-$-mixing property.

Summary updated

Abstract

Let {X n , n 1} be a strictly stationary r --mixing sequence of positive random
BibTeX
@article{tan2012almost,
  author = {Xili Tan and Ying Zhang and Yong Zhang},
  title = {{An almost sure central limit theorem of products of partial sums for ρ--mixing sequences}},
  journal = {Journal of Inequalities and Applications},
  year = {2012},
  doi = {10.1186/1029-242x-2012-51},
}
Figure