scientific article; zbMATH DE number 3862282

From MaRDI portal
Publication:3330345





[For the entire collection see Zbl 0536.00013.] Consider the simple equation (of n-dim.) \(dX_ t=\sum a^ o_ j\phi_ j(X_ t)dt+dB_ t\) where B is the (n-dim.) Wiener process, \(\phi_ j\) are given functions, while \((a^ o_ j)\) are the parameters to be estimated. If the system has the unique invariant density \(f_ 0\), and for a solution process X, \(\mu(X_ t)=\sum a^ o_ j\phi_ j(X_ t)\in L^ 2,\) then the maximum likelihood estimator (up to time T) \(\hat a{}_{ML}(T)\) converges (a.e.) to \(a^ o\), and the central limit theorem holds for \(\hat a{}_{ML}(T)\). The estimator \(\hat a{}_{ML}(T)\) is to maximize \[ L(a;T)=\exp \{-2^{- 1}\int^{T}_{0}(\sum a_ j\phi_ j(X_ t))^ 2dt+\int^{T}_{0}(\sum a_ j\phi_ j(X_ t))dX_ t\} \] and is given by \(\hat a{}_ M(T)=C^{-1}\rho\), where \(c_{ij}\equiv \int^{T}_{0}\phi '\!_ i(X_ s)\phi_ j(X_ s)ds,\) and \(\rho_ j=\int^{T}_{0}\phi '\!_ j(X_ s)dX_ s, \phi\) ' being the transpose of \(\phi\).











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3330345)