What's the limiting distribution of the eigenvalue and eigenvectors of the sample covariance matrix?

68 Views Asked by At

The question is from the book 'Statistical Models and Methods for Financial Markets', Page 43.
Suppose $\boldsymbol{x_1,..., x_n}$ are $n$ independent observations from a multivariate population with mean $\boldsymbol{μ}$ and covariance matrix $\boldsymbol{V}$
The mean vector μ can be estimated by $$\bar{\boldsymbol{x}} = \sum_{i=1}^n \boldsymbol{x}_n$$ And the covariance matrix can be estimated by $$ \hat{\boldsymbol{V}} = \sum_{i=1}^n (\boldsymbol{x}_i - \bar{\boldsymbol{x}}) (\boldsymbol{x}_i - \bar{\boldsymbol{x}})^T/(n-1) $$

which is the sample covariance matrix.
Let $\hat{\boldsymbol{a}}_j = (\hat{a}_{1j} ,..., \hat{a}_{pj} )^T$ be the eigenvector corresponding to the $j$th largest eigenvalue $\hat{\lambda_j}$ of the sample covariance matrix $\hat{\boldsymbol{V}}$.
For fixed $p$, the following asymptotic results have been established under the assumption $λ1 > ··· > λp > 0$ as $n\to \infty$:
(i) $\sqrt{n} (\hat{\lambda}_i -\lambda_i)$ has a limiting $N(0, 2\lambda_i^2 )$ distribution.
Moreover, for $i \neq j$, the limiting distribution of $\sqrt{n} (\hat{\lambda}_i -\lambda_i,\hat{\lambda}_j -\lambda_j)$ is that of two independent normals.
(ii) $\sqrt{n} (\hat{\boldsymbol{a}}_i -\boldsymbol{a}_i)$ has a limiting normal distribution with mean $\boldsymbol{0}$ and covariance matrix $$ \sum_{h\neq i} \frac{\lambda_i \lambda_h}{(\lambda_i - \lambda_h)^2} \boldsymbol{a}_h \boldsymbol{a}_h^T $$
It seems like it's a direct application of multivariate CLT,but I cannot derive these two statements.