Strict local martingale deflators and valuing American call-type options (Q1761442): Difference between revisions

Let \(\beta\) be a strictly positive and nonincreasing process such that \(\beta_0= 1\), and let \(S\) be a strictly positive semimartingale. Furthermore, let \(Z\) be a strictly positive local martingale with \(Z_0= 1\) (all processes being defined on a filtered probability space \((\Omega,({\mathcal F}_t)_{t\geq 0},\operatorname{P})\), where \(\operatorname{P}\) is defined on \((\Omega,{\mathcal F}_\infty)\) given by \({\mathcal F}_\infty= \bigvee_{t\geq 0}{\mathcal F}_t\), the filtration satisfying the usual hypotheses). Assume that \(L:= Z\beta S\) is a local martingale. Let \(g: \mathbb{R}_+\to \mathbb{R}_+\) be a nonnegative convex function such that \(g(0)= 0\), \(g(x)< x\) for all \(x>0\) and \(g(x)/x\to 1\) \((x\to\infty)\). Set \(X:= Yg(S)\) and assume that \(X_\infty:= \lim_{t\to\infty} X_t\) exists. Consider the optimization problem (OS): compute the value \(v:= \sup_{\tau\in{\mathcal T}} \operatorname{E}[X_\tau]\); find \(\widehat\tau\in{\mathcal T}\) such that \(\operatorname{E}[X_{\widehat\tau}]= v\) (\({\mathcal T}\) denoting the family of all \(({\mathcal F}_t)\)-stopping times). A stopping time \(\widehat\tau\in{\mathcal T}\) is called optimal if \(\operatorname{E}[X_{\widehat\tau}]= v\). Let \(\tau\in{\mathcal JT}\) and let \((\sigma^n)\) be a localizing sequence for \(L\). The authors show that the sequence \((\operatorname{E}[X_{\tau\wedge\sigma^n}])\) \((n\geq 1)\) is nondecreasing. Furthermore, it is shown that \(\delta(\tau):= \lim_{n\to\infty} \operatorname{E}X_{\tau\wedge\sigma^n}]- \operatorname{E}[X_\tau]\) (being nonnegative) does not depend on the localizing sequence \((\sigma^n)\) for \(L\). Finally, a certain stopping time \(\tau^*\) is defined which -- combined with \(\delta(\cdot)\) -- allows for the following complete solution of (OS). 1. The value of (OS) is \(v= \operatorname{E}[X_{\tau^*}]+ \delta(\tau^*)= \operatorname{E}[X_\infty]+ \delta(\infty)\). 2. \(\widehat\tau\in{\mathcal T}\) is optimal iff \(\tau^*\leq\widehat\tau\) as well as \(\delta(\widehat\tau)= 0\). 3. Optimal stopping times exist iff \(\delta(\tau^*)= 0\). In that case, \(\tau^*\) is the smallest optimal stopping time, and the set of all optimal stopping times equals \(\{\widehat\tau\in{\mathcal T}: \tau^*\leq\widehat\tau\) and \(\delta(\widehat\tau)= 0\}\). This solves an open question by \textit{I. Karatzas} and \textit{R. Fernholz} [in: A. Bensoussan (ed.) et al., Handbook of numerical analysis. Vol XV. Special Volume: Mathematical modeling and numerical methods in finance. Amsterdam: Elsevier/North-Holland. Handbook of Numerical Analysis 15, 89--167 (2009; Zbl 1180.91267)].