1: \begin{abstract}

2: 	We propose a novel Metropolis-Hastings algorithm to sample uniformly from

3: 	the space of correlation matrices. Existing methods in the literature are

4: 	based on elaborated representations of a correlation matrix, or on complex

5: 	parametrizations of it. By contrast, our method is intuitive and simple,

6: 	based the classical Cholesky factorization of a positive definite matrix and

7: 	Markov chain Monte Carlo theory. We perform a detailed convergence analysis

8: 	of the resulting Markov chain, and show how it benefits from fast

9: 	convergence, both theoretically and empirically. Furthermore, in numerical

10: 	experiments our algorithm is shown to be significantly faster than the

11: 	current alternative approaches, thanks to its simple yet principled

12: 	approach.

13: 	\keywords{Correlation matrices \and Random sampling \and Metroplis-Hastings}

14: \end{abstract}

15: