Particle-based online estimation of tangent filters with application to parameter estimation in nonlinear state-space models (Q2304257)

On the basis of the approach taken by \textit{P. Del Moral} and \textit{A. Guionnet} [Ann. Inst. Henri Poincaré, Probab. Stat. 37, No. 2, 155--194 (2001; Zbl 0990.60005)], the authors developed an efficient and numerically stable algorithm for particle approximation of the filter derivatives in SSMs (State-Space Models). The proposed approach includes, as a sub-routine, the particle-based rapid incremental smoother (short PaRIS), introduced by the same authors in [Bernoulli 23, No. 3, 1951--1996 (2017; Zbl 1392.62252)], an algorithm that is tailor-made for online approximation of smoothed additive state functionals. This result is analogous to the results obtained by \textit{T. N. M. Nguyen} et al. [Adv. Appl. Probab. 50, No. 1, 154--177 (2018; Zbl 07163641)] for the case of particle-based two-filter smoothing. Moreover, by assuming strong mixing of the model, the asymptotic variance can be bounded uniformly in time. The numerical performance of this online parameter estimation algorithm is investigated in a simulation study, and seems to obtain significant improvement over the approach of \textit{P. Del Moral} and \textit{A. Guionnet} [Ann. Inst. Henri Poincaré, Probab. Stat. 37, No. 2, 155--194 (2001; Zbl 0990.60005)] for online parameter learning in the stochastic volatility model given by \textit{J. Hull} and \textit{A. White} [in: Stochastic volatility. Selected readings. Oxford: Oxford University Press. 109--129 (2005; Zbl 1126.91369)].