Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I need to find an example of an Ito process $X=\{X_t:t\in[0,T]\}$ with non-zero Ito integral part and a sequence of Ito processes $\{X_n\}$ such that $X_n$ converges uniformly to $X$, as $n\to\infty$, and $[X_n]_T\to \frac{1}{2}[X]_T$, as $n\to\infty$. Here $[X]$ denotes the quadratic variation process of an Ito process $X$. I have no idea how to start on this one so I would like a hint on a direction.

share|improve this question
add comment

1 Answer

up vote 1 down vote accepted

Fix $a$ in $(0,1)$ and a standard Brownian motion $X$. For every subdivision $\sigma=(t_k)_{0\leqslant k\leqslant n}$ of $[0,T]$, deterministic or random, consider the diffusion $X^\sigma$ defined by $X^\sigma_0=0$ and $$ \mathrm dX^\sigma_t=a\mathrm dX_t+(1-a)b_\sigma(t)\mathrm dt, $$ with $$ b_\sigma(t)=\sum\limits_{k=1}^n\frac{X_{t_k}-X_{t_{k-1}}}{t_k-t_{k-1}}\cdot[t_{k-1}\leqslant t\lt t_k]. $$ In words $X^\sigma_{t}=X_{t}$ for every $t$ in $\sigma$ and, on every interval $[t_{k-1},t_k]$ of $\sigma$, a proportion $1-a$ of the variation of $X$ is transformed into a drift which performs a linear interpolation from $X_{t_{k-1}}$ to $X_{t_k}$. In particular, for every $t$ in $[0,T]$, $$ [X^\sigma]_t=a^2[X]_t=a^2t. $$ To estimate the uniform norm of $X^\sigma-X$ on $[0,T]$, first note that this is the maximum of the uniform norms of $X^\sigma-X$ on the intervals of $\sigma$ and that the distribution of these depends only on the length of the interval. The uniform norm on an interval of length $s$ is distributed like $(1-a)M_s$ with $$ M_s=\max\limits_{0\leqslant t\leqslant s}|X_t-(t/s)X_s|. $$ Fix a positive $x$. Then $\mathrm P(M_s\geqslant x)$ behaves roughly like $\mathrm e^{-x^2/(2s)}$ when $s\to0^+$.

Consider a subdivision $\sigma$ with $n$ points and with step $s$. By the union bound, the uniform norm of $X^\sigma-X$ on $[0,T]$ is at least $(1-a)x$ with probability at most $n\mathrm e^{-x^2/(2s)}$. For the regular subdivision $\sigma_n$ with $n$ intervals, $s=T/n$ hence the probability is at most $n\mathrm e^{-nx^2/(2T)}$.

The series $\sum\limits_nn\mathrm e^{-cn}$ is summable for every $c\gt0$ hence, for every $x\gt0$, $\|X-X^{\sigma_n}\|\leqslant x$ for every $n$ large enough, almost surely. This proves that $X^{\sigma_n}\to X$ uniformly, almost surely.

share|improve this answer
    
I assume that the expression in the brackets in the definition of $b_{\sigma}$ mean an indicator function. I get that $X^{\sigma}_{t_k}=X_{t_k}-(1-a)X_{t_{k-1}}$, or am I wrong? This is an interesting example, but maybe one can use some fact to construct a more straightforward one? –  nokiddn Mar 18 '12 at 12:22
    
Yes, this is Iverson bracket. // No, $X^\sigma_t=X_t$ for every $t$ in $\sigma$. –  Did Mar 18 '12 at 12:24
    
Sorry, I still don't get it. This is how i proceed: let $t_k\in\sigma$, then $$X^{\sigma}_{t_k}=aX_{t_k}+(1-a)\int^{t_k}_{t_{k-1}}\frac{X_{t_k}-X_{t_{k-1}}}{‌​t_k-t_{k-1}}ds=X_{t_k}-(1-a)X_{t_{k-1}}.$$ Where is my mistake? –  nokiddn Mar 18 '12 at 12:34
    
In the fact that the integral should start at $s=0$ and not at $s=t_{k-1}$, yielding the value $\sum\limits_{i=1}^k(X_{t_i}-X_{t_{i-1}})=X_{t_k}$. –  Did Mar 18 '12 at 12:40
    
Thank you, I mistakenly kept only one member of the sum. –  nokiddn Mar 18 '12 at 13:01
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.