A method of second order accuracy integration of stochastic differential equations

For the stochastic differential equation
$$ dX=a(t,X)\,dt+\sigma(t,X)\,dw,\qquad X(t_0)=x,\ t_0\le t\le t_0+T, $$
the problem of approximate calculation of the expectation $\mathbf Mf(X_{t_0,x}(t_0+T))$ is considered.
Rather a simple method is proposed for recursive modeling of random variables
$$ \overline X_{t_0,x}(t_k);\quad k=0,1,\dots;\quad t_k=t_0+kh;\quad h=\frac{T}{m}; $$
such that
$$ \mathbf Mf(X_{t_0,x}(t_0+T))=\mathbf Mf(\overline X_{t_0,x}(t_0+T))+O(h^2). $$