自回归模型
数学
应用数学
渐近分析
样品(材料)
协方差矩阵
基质(化学分析)
协方差
相(物质)
协方差矩阵的估计
相变
随机矩阵
渐近分析
估计理论
订单(交换)
统计物理学
统计
作者
Tiefeng Jiang,Tuan Pham
摘要
We investigate the asymptotic distribution of the largest off-diagonal entry of the sample covariance matrix under the autoregressive structure driven by a parameter r=rn∈(0,1). Our analysis reveals a phase transition at the order of logp/n in the asymptotic distribution of the largest entries when r is bounded away from 1. Additionally, as r converges to 1 with logarithmic rate, we show that there is a subtle clustering effect among the entries, leading to an asymptotic distribution characterized by a Gumbel law with an extremal index strictly between 0 and 1. This unexpected behavior is counterintuitive and represents a significant departure from the standard Gumbel law observed in (Ann. Statist. 53 (2025) 907–928) for sample correlation matrices: the seemingly simpler sample covariance matrices can exhibit more complex behavior than the sample correlation matrices. Our proof methodology relies on high-dimensional Gaussian approximation techniques and employs a blocking argument to create independence.
科研通智能强力驱动
Strongly Powered by AbleSci AI