线性模型中误差方差估计的指数收敛速度

Suppose given a linear model y_j=x'_jβ+μ_j, j=1,2,…, The random errors all have a mean zero and unknown variance σ², 0<σ²<∞. Let σ_n² be the estimate of σ² based on the residual sum of squares and calculated from (x_j, y_j), j=1,…,n. In this paper we show that if μ₁,μ₂,…, are independent but not necessarily identically distributed, and some further conditions on {μ_j} and (x₁|…|x_n) are satisfied, then for any ε>0 there exist constant ρ_ε, 0<ρ_ε<1, Such that P(|σ_n²-σ²|≥ε)=O(ρ_εⁿ).

Exponential Convergence Rates of Error-variance Estimates in Linear Models