Non‐parametric Kernel Estimation of the Coefficient of a Diffusion (Q4956067)

The diffusion process \(X_t\), defined by \(dX_t=b_tdt+\sigma(X_t)dw_t\), is observed at the points \(t=i/n\), \(i=1,\dots,n\). The author proposes an estimator \(\Gamma_n(x)\) of \(\sigma(x)\) at points \(x\) at which the process spends enough time. The estimator uses Baxter-type convergence. Main conditions of \(\Gamma_n(x)\) consistency are the following: (i) For some \(r\geq 1\), \(\sigma(x)\) is \([r]\) times continuously differentiable, non-vanishing and the \([r]\)-th derivative is locally Hölder with index \(r-[r]\); (ii) The process \(Y_t\) defined by \(dY_t=\sigma(Y_t)dw_t\) does not explode in a finite time; (iii) The process \(b_t\) is such that the laws of \(X\) and \(Y\) are locally equivalent. Under these assumptions \(\Gamma_n(x)-\sigma^2(x)=O(n^{-r(1+2r)})\) and analogous convergence rates in \(L_p\) are obtained.