Embedding laws in diffusions by functions of time (Q888534): Difference between revisions

Consider a recurrent diffusion \((X_t)_{t\geq 0}\) on the real line started at \(0\), e.g., a standard Brownian motion, and let \(\mu\) denote a given probability measure on \(\mathbb{R}\) with \(\mu(\{0\})= 0\). Then, there exist unique left-continuous functions \(b: (0,\infty)\to\mathbb{R}\cup \{\infty\}\) increasing and \(c: (0,\infty)\to\mathbb{R}\cup\{-\infty\}\) decreasing such that \(X_\tau\) with \(\tau:= \text{inf}\{t>0\mid X_t\geq b(t)\) or \(X_t\leq c(t)\}\) is distributed according to \(\mu\). In their proof, the authors use weak convergence arguments arising from Helly's selection theorem. The proof is constructive, i.e., a simple algorithm is supplied for computing sequences of functions approximating \(b\) and \(c\), respectively. Furthermore, suppose that \(b=+\infty\) if \(\text{supp}(\mu)\subseteq\mathbb{R}^-\) and \(c=-\infty\) if \(\text{supp}(\mu)\subseteq\mathbb{R}^+\). Then, if \(\sigma\) is any stopping time such that \(X_\sigma\sim X_\tau\), \(\tau\) is optimal, i.e., if \(F: \mathbb{R}^+\to\overline{\mathbb{R}}\) is concave such that \(\mathbb{E} F(\tau)\) exists, then \(\mathbb{E} F(\tau)\leq\mathbb{E} F(\sigma)\), in particular \(\mathbb{E}(\tau\wedge T)\leq\mathbb{E}(\sigma\wedge T)\) for all \(T>0\). If, in addition \(\sigma\leq\tau\) a.s., then even \(\sigma=\tau\) a.s., i.e., \(\tau\) is minimal in the sense of \textit{I. Monroe} [Ann. Math. Stat. 43, 1293--1311 (1972; Zbl 0267.60050)]. As a corollary, \((X_{t\wedge\tau})_{t\geq 0}\) satisfies natural uniform integrability conditions in terms of \(\mu\).