Nonparametric inference for a doubly stochastic Poisson process (Q2366192)

Let the two-dimensional process \(\bigl(n(t),z(t)\bigr)\) be observed in the time interval \((0,T)\). It is assumed that \(n(t)\) and \(z(t)\) are, respectively, counting and predictable. Moreover, they are related by \[ n(t)= \int^ t_ 0 a\bigl(z(s)\bigr) ds+ m(t), \] where \(a(x)\) is a nonnegative nonrandom function and \(m(t)\) is a zero-mean local martingale. Further, let \(A(x)= \int^ x_ 0 a(u)du\). The author suggests estimators for the two functions \(a(x)\) and \(A(x)\) assuming \(0\leq x\leq 1\). As estimators the following statistics are suggested: \[ A^*(x)= n^{-1} \sum^{[xn]}_{k=1}\left(\int^ T_ 0 I((k- 1)/n<z(t)\leq k/n)dn(t)\Bigl/\int^ T_ 0 I((k-1)/n< z(t)\leq k/n)dt\right) \] \[ \text{and}\quad a^*(x)= h^{-1} \int K\bigl((x- y)/h\bigr) A^*(dy). \] Here \([x]\) is the integer part of \(X\), \(K(\cdot)\) is a kernel function, and it is assumed that \(T\to\infty\), \(n=n_ T\to\infty\) and \(h=h_ T\to 0\). Under certain conditions the convergence of \(A^*(x)- A(x)\) in Skorokhod topology to a Gaussian process is established. A number of applications and statistical inferences are also given.