On the free boundary of an annuity purchase (Q1711720)

A customer has an investment in a risky asset modelled by a geometric Brownian motion. The customer consumes continuously a portion \(\alpha X_t\; d t\) of the wealth \(X_t\) at time \(t\). At the time of death \(\Gamma\), the remaining wealth is consumed. At a time \(\tau \in [0,T]\), the customer has the possibility to convert the wealth into a life-long annuity, calculated fairly using the first order basis. There is a deadline for the conversion if \(T < \infty\). The problem is to maximise \[ V_t = \sup_{\tau \in T_{t,T}} E^{\mathrm{S}}\Bigl[\int_t^{\Gamma\wedge \tau} e^{-\rho s} \alpha X_s\;d s + \mathrm{I}_{\Gamma \le \tau}e^{-\rho\Gamma} X_\Gamma + P_\tau \int_{\Gamma\wedge \tau}^{\Gamma} e^{-\rho s} \;d s \Bigm| \mathcal{F}_t\Bigr] \] on \(\{\Gamma > t\}\). Here, \(E^{\mathrm{S}}\) denotes the probability measure with the mortality the customer believes in, which may be different from the first order basis, and \(P_\tau\) is the annuity rate. The problem can be formulated in the form \[ V_t = \sup_{\tau \in T_{t,T}} E\Bigl[\int_t^{\tau} e^{-\int_t^s r(u) \;d u} \beta(s) X_s\;d s + e^{-\int_t^\tau r(u) \;d u} G(\tau,X_\tau) \Bigm| \mathcal{F}_t\Bigr]\;, \] where \(E\) is the measure under which the wealth process is defined. Considering \(v(t,x) = V(t,x) - G(t,x)\) with \(V(t,x) = V_t\), the problem reads \[ v(t,x) = \sup_{0 \le \tau \le T-t} E\Bigl[\int_0^{\tau}e^{-\int_0^s r(t+u) \;d u} H(t+s,X_s^x)\;d s\Bigr]\;, \] where \(X_s^x\) is the wealth process starting at \(X_0^x = x\). This is now a classical optimal stopping problem. However, with the difficulty that the value function is not monotone. It is shown that, under some assumptions, the stopping and the continuation region are split by a locally Lipschitz boundary function \(b(t)\). A verification theorem is proved and an equation is given, from which the free boundary can be calculated. A numerical example illustrates the findings.