I have a conjecture that seems rather obvious but the proof seems elusive.

Consider a diffusion given by, $dX_t = \mu(X_t) dt + \sigma(X_t) db_t$

where $b_t$ is a standard Brownian motion.

$\mu,\sigma$ are piecewise Lipschitz, bounded, of locally bounded variation. If there are any other nice properties required, I would assume I have them.

Suppose we are interested in computing the following -

$v(x) = \mathbb{E}^x \int_0^\infty e^{-rt} f(X_t) dt$

where $f$ is the flow payoff that our agent collects. $f$ is also piecewise Lipschitz, bounded, locally bdd variation etc. Let $f$ be between $m$ and $M$.

I want to prove that $v$ is continuous. The reason why I think it sounds obviously true is the following -

Suppose we want to prove continuity at $x$. Fix some point $a$ to the left of $x$ and consider a sequence $\{x_n\} \in [a,x]$ such that $x_n \rightarrow x$.

Let $\tau_n := \inf \{t: X^{x_n}_t \notin (a,x)\}$


$v(x_n) \le \mathbb{E}^{x_n} \left[ e^{-r\tau_n} [ P(X^{x_n}_{\tau_n} = a) v(a) + P(X^{x_n}_{\tau_n} = x) v(x)] + (1-e^{-r\tau_n})M\right]$

$M$ is because that's the upper bound of $f$.

Now, if, $x_n \rightarrow x$ $\Rightarrow$ $\tau_n \rightarrow 0$, and $P(X^{x_n}_{\tau_n} = x) =1$.

then we can get continuity. But I do not know how to prove these two things which sound so obviously true. The only approach I am aware of is the time change formula but that seems like an overkill for this problem. Any help would be immensely appreciated. Thanks.

  • $\begingroup$ Three comments: (1) You obviously need some lower bound on $|\sigma(x)|$. (2) Strictly speaking, your argument only gives upper semi-continuity, you also need a simpler bound $\nu(x_n) \ge \mathbb{E}^{x_n}[e^{-r\tau_n}[P(X_{\tau_n}^{x_n}=a)]]\nu(a)$. (3) Your process is "locally Feller" except at a discrete set of points, so you only need to worry about points where any of the coefficients is discontinuous. $\endgroup$ – Mateusz Kwaśnicki Aug 28 '17 at 8:49

The key thing is that your diffusion is Feller under the stated conditions, so $x \mapsto \mathbf E_x f(X_t)$ is continuous. Therefore $x \mapsto \int_0^{\infty} \mathbf E_x f(X_t) e^{-rt} \ d t$ is continuous and using Fubini gives the required result. For reference material on continuous dependence on the initial condition see e.g. Karatzas and Shreve, Rogers & Williams, or Kallenberg.

| cite | improve this answer | |
  • $\begingroup$ Thanks for the answer. From what I know, the diffusion is not Feller continuous because the drift and volatility are piecewise Lipschitz and not Lipschitz everywhere. Also, $f$ is piecewise Lipschitz. Do you think there are some standard results like the one you are referring to that can do our job? $\endgroup$ – avk255 Jun 3 '16 at 13:44
  • $\begingroup$ Are $\mu$ and $\sigma$ locally Lipschitz and in particular continuous? In this case, further assuming you have uniqueness of the martingale solution, Kallenberg Thm 21.11 gives the Feller property. $\endgroup$ – Joris Bierkens Jun 3 '16 at 13:46
  • $\begingroup$ No, $\mu$ and $\sigma$ are piecewise continuous. We can assume that the number of discontinuities is finite. They are bounded, $\sigma$ is bounded from below by a strictly positive number. Hence, by Nakao, I will have a strong solution. But, I must work with piecewise continuous $\mu$ and $\sigma$. $\endgroup$ – avk255 Jun 3 '16 at 13:55
  • $\begingroup$ How about $f$, is it not continuous either? $\endgroup$ – Joris Bierkens Jun 3 '16 at 14:16
  • $\begingroup$ $f$ is not continuous either. It is piecewise continuous, bounded, finite points of discontinuity. $\endgroup$ – avk255 Jun 3 '16 at 17:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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