1: \begin{abstract}

2: We study concentration in spectral norm of nonparametric estimates of correlation matrices. 

3: We work within the confine of a Gaussian copula model. 

4: Two nonparametric estimators of the correlation matrix, 

5: the sine transformations of the Kendall's tau and Spearman's rho correlation coefficient, are studied. 

6: Expected spectrum error bound is obtained for both the estimators. 

7: A general large deviation bound for the maximum spectral error of a collection of 

8: submatrices of a given dimension is also established. 

9: These results prove that when both the number of variables and sample size are large, 

10: the spectral error of the nonparametric estimators is of no greater order than that of the latent sample 

11: covariance matrix, at least when compared with some of the sharpest known error bounds for the later. 

12: As an application, we establish the minimax optimal convergence rate in the estimation of high-dimensional 

13: bandable correlation matrices via tapering off of these nonparametric estimators. 

14: An optimal convergence rate for sparse principal component analysis is also established 

15: as another example of possible applications of the main results.  

16: \end{abstract}

17: