6

Suppose I have an $n$-dimensional Itô SDE $$dX_t = \sigma(X_t) dW_t + \lambda(X_t)dt$$ and I'm interested in diffusion bridges from $X_0=a\in\mathbb R^n$ to $X_T=b\in\mathbb R^n$.

Now let $Y_t$ be a diffusion bridge of the SDE $$dY_t=\sigma(Y_t)dW_t - \lambda(Y_t) dt$$ from $Y_0=b$ to $Y_T=a$.

Based on a (very) non-rigorous approximation argument I would conjecture that the laws of $X_t$ and $Y_{T-t}$ agree.

To make my proof rigorous would involve $\varepsilon$'s and $\delta$'s that I think would quite long. Is this a known result? Is there a proof using a theorem about diffusions. Or is it wrong?

Pointers etc are very much appreciated.

Tim
  • 5,564
  • 1
    Note: as you've written it, they are not independent processes (same Brownian motion $W_t$). Is it not sufficient to just point out that every path of $Y_t$ corresponds (bijectively) to a path of $X_t$ by reversing time? (After time reversion it is the same SDE.) – Kirill Jul 18 '13 at 00:31
  • @Kirill Thanks for the comment. You're right, I think that is sufficient. I knew it would be something obvious like that that I'd missed. – Tim Jul 18 '13 at 07:49
  • @Kirill I'm unsure how the detail of this works. Would you mind posting a proof? Thanks! – Ben Derrett Jul 20 '13 at 12:09

1 Answers1

3

The laws of the processes do not agree. Time-reversal for stochastic processes is more complicated than just to swich signs in the drift compnent. Think of an Ornstein-Uhlenbeck process started from an equilibrium for example. Just google for time-reversed diffusions for more help.

Eulsen
  • 31