1: \begin{abstract}

2: We consider the identification of large-scale linear and stable dynamic systems

3: whose outputs may be the result of many correlated inputs.

4: Hence, severe ill-conditioning may affect the estimation performance.%problem.

5: This is a scenario often arising when modeling complex physical systems

6: given by the interconnection of many sub-units where feedback

7: and algebraic loops can be encountered.

8: We develop a strategy based on Bayesian regularization

9: where any impulse response is modeled as the realization of a zero-mean Gaussian process.

10: The stable spline covariance is used to include information

11: on smooth exponential decay of the impulse responses.

12: We then design a new Markov chain Monte Carlo scheme

13: that deals with collinearity and is able to efficiently reconstruct the posterior of the impulse responses.

14: It is based on a variation of Gibbs sampling which updates possibly overlapping blocks of the parameter space

15: on the basis of the level of collinearity affecting the different inputs.

16: Numerical experiments are included to test the goodness of the approach

17: where hundreds of impulse responses form the system and inputs correlation may be very high.

18: %We also discuss theoretical questions underlying the convergence rate

19: %of the proposed sampler.

20: \end{abstract}

21: