Central limit theorem for a Stratonovich integral with Malliavin calculus (Q359692)

This article derives a change of variable formula for Stratonovich type integral processes driven by a Gaussian process that is not necessarily a semimartingale. The central technical ingredient is the multidimensional extension of a central limit theorem for iterated Skorokhod integrals obtained by \textit{I. Nourdin} and \textit{D. Nualart} in [J. Theor. Probab. 23, No. 1, 39--64 (2010; Zbl 1202.60038)]. More precisely, they consider the limit of the ``midpoint'' Riemann sums \[ \Phi_n(t) := \sum_{j=1}^{\lfloor nt/2\rfloor} f'(W_{(2j-1)/n)}\left( W_{2j/n}-W_{(2j-2)/n} \right), \] where \(f\) is a suitable function and \(W=(W_t)_{t\geq 0}\) is a Gaussian process with a technical assumptions on the covariance function. The authors show that the sequence \((W_t,\Phi_n(t))_{t\geq 0}\) has a limit \((W_t,\Phi(t))_{t\geq0}\) in distribution on the Skorokhod space \(\mathbb{D}^2[0,\infty)\) and that the second component is interpreted as Stratonovich integral \(\Phi(t)=\int_0^tf(W_s)\circ dW_s\). They derive a change of variable formula for the Stratonovich integral of the form \[ f(W_t) = f(W_0) + \int_0^t f'(W_s)\circ dW_s + \frac{1}{2}\int_0^t f''(W_s)dB_s. \] In addition to the usual Leibniz rule of the Stratonovich calculus for continuous martingales, an Itō integral of the second derivative appears. Here, \(B=(B)_{t\geq 0}\) is a scaled Brownian motion independent of \(W\), with variance \(\operatorname{E}B_t^2 =2\eta(t)\), where \(\eta(t)\) is determined by the covariance structure of \(W\). In the example of a bifractional Brownian motion with parameter \(H\leq 1/2\), \(H\times K =1/4\), the scaling function \(\eta\) is calculated explicitly. The article is structured as follows. After a brief introduction to the Malliavin calculus of isonormal Gaussian processes, the authors proof a multidimensional version of the central limit theorem of Nourdin and Nualart [loc. cit.] stating that iterated Skorokhod integrals \(F_n= \delta^q(u_n)\) converge stably and that the limit has a conditional Gaussian distribution \(\mathcal{N}(0,\Sigma)\) given \(X\). The main step is then to identify the second-order terms in a Taylor expansion of \(f\) as Skorokhod integrals of indicator functions of the form \[ \begin{multlined} f''(W_{(2j-1)/n})\left( \Delta W^2_{2j/n}-\Delta W^2_{(2j-1)/n} \right)\\ = f''(W_{(2j-1)/n})\delta^2 \left(\mathbf{1}_{[(2j-1)/n,j/n]}^{\otimes 2}-\mathbf{1}_{[(2j-2)/n,(2j-1)/n]}^{\otimes 2} \right),\end{multlined} \] where the authors apply their multidimensional version of the central limit theorem for Skorokhod integrals, generating the additional Gaussian term.