Page 438 - Special Topic Session (STS) - Volume 3
P. 438

STS555 Patrice Bertail et al.



                                   Robust estimators for some piecewise-
                                       deterministic markov Processes
                                              1
                                                               2
                                                                             2
                                 Patrice Bertail  ,Gabriela Ciolek  ,Charles Tillier
                                             1  Universit´e Paris Nanterre
                                                2 TelecomParisTech

                  Abstract
                  This talk is devoted to extending the notion of robustness to Markov chains
                  with applications to PDMP, based on their (pseudo-) regenerative properties.
                  Precisely, it is shown how it is possible to define the "influence function" in this
                  framework, so as to measure the impact of (pseudo-) regeneration data blocks
                  on the statistic of interest. We establish some asymptotic results for robust
                  estimators of some usual functionals of the stationary measure. We also define
                  the concept of regeneration-based signed linear rank statistic and L-statistics,
                  as specific functionals of the regeneration blocks, which can be made robust
                  against outliers in this sense. Indeed, even the usual quantile is not robust in
                  this  framework.  We  essentially  apply  these  notions  to  reservoir  models  in
                  insurance and hydrology. We obtain robust estimators of probability of ruins
                  and quantiles (value at risk).

                  Keywords
                  Markov Chains; Piecewise Deterministic Markov processes; influence function

                  1.  Regenerative Markov chains: Notations and context
                      Denote by X = (Xn)n∈N a positive recurrent Markov chain on a countably
                  generated state space (E, ) with transition probability Π and initial probability
                  ν. For any B ∈ E and n ∈ N, we have

                               X 0 ∼ ν and P(X n+1 ∈ B|X 0, · · ·, X n) = Π(X n, B) a.s.

                  In the following, Px (resp. Pν) designates the probability measure such that X0
                  = x and X0 ∈ E (resp. X0 ∼ ν), and Ex (·) is the Px-expectation (resp. Eν (·) is the
                  Pν-expectation). All along this paper, we suppose that X is ψ -irreducible and
                  aperiodic Markov chain. We are particularly interested in the atomic structure
                  of Markov chains as in [1].

                  Definition 1. Suppose that X is aperiodic and ψ-irreducible. We say that a set
                  A ∈  is an accessible atom if for all x, y ∈ A we have (x, ·) = Π(y, ·) and ψ(A) >
                  0. In that case we call X atomic.







                                                                     427 | I S I   W S C   2 0 1 9
   433   434   435   436   437   438   439   440   441   442   443