Hiding a drift (Q971947)

Let \(B= (B_t)\) be a standard Brownian motion (BM) on some probability space \((\Omega,{\mathcal F},P)\). For a fixed constant \(\mu> 0\) let \(S_t= \mu_t+ B_t\) \((t\geq 0)\) denote the BM with drift \(\mu\). Let \(P^B:= ({\mathcal F}^B_t)\) \((t\geq 0)\) denote the right-continuous saturated filtration generated by \(B\). Given a predictable, \(F^B\)-adapted process \(H= (H_t)\) \((t\geq 0)\) let \((H\cdot S)= ((H\cdot S)_t)\) \((t\geq 0)\) be the stochastic integral in its right-continuous, saturated filtrations \(F^{(H\cdot S)}:=({\mathcal F}^{H\cdot S)}_t)\) \((t\geq 0\). The authors are dealing with the following Question 1 (due to M. Yor): Can we define an \(F^B\)-predictable process \(H\) such that \((H\cdot S)\) is a BM (without drift) in its own filtration, i.e. an \(F^{(H\cdot S)}\)-BM? Although Question 1 still remains an open problem, the authors obtain the following Theorem. Let \(B\) be a standard BM and let \(F^B\) be as above. Then (i) For each \(\mu> 0\) there are \(F^B\)-predictable processes \(H= (H_t)\) \((t\geq 0)\) taking values in \(\{-1,1\}\) and \((\mu_t)\) \((t\geq 0)\) taking values in \((0,2\mu)\) such that for the process \(S= (S_t)\) given by \(dS_t= \mu_t dt+ dB_t\), \(S_0= 0\), we have that \((H\cdot S)\) is a BM in its own filtration J\(F^{(H\cdot S)}\). (ii) Furthermore, for each \(\delta> 0\) we can choose \((\mu_t)\) such that \(\mu-\delta< \mu_t< \mu+\delta\) for all \(t\geq 0\). Additionally, the authors obtain a positive solution of Question 2 (due to M. Emery): Let \((\varepsilon_n)\) \((n\leq 0)\) be an i.i.d. sequence in its natural filtration \({\mathcal F}_n)\) \((n\leq 0)\) such that \[ P(\varepsilon_n= 1)= 1- P(\varepsilon_n=-1):= p\in (0,1)\setminus \{1/2\}. \] Is there an \(({\mathcal F}_n)\)-predictable sequence \((h_n)\) \((n\leq 0)\) of \(\{-1,1\}\)-valued random variables such that \((h_n\varepsilon)\) \((n\leq )\) is i.i.d. with \(P(h_n\varepsilon_n= 1)= P(h_n\varepsilon_n=-1)= 1/2\)?