Almost sure optimal hedging strategy (Q2511561)

The authors find an asymptotic lower bound in almost sure sense, for the renormalized quadratic variation of the Hedging error in discrete hedging using stopping times, for a large class of payoffs, under a \(d\)-dimensional Itō price model. Moreover they find an explicit strategy that attains this lower bound. More explicitly, consider a price process \(S.\) Without losing generality, in the paper it is assumed null interest rate and null drift. For simplicity, I summarize in this review only the one-dimensional case. Therefore, \(S\) is a continuous local martingale \[ S_t=S_0+\int_0^t \sigma_s dB_s, \,\quad t\in [0,T], \] where \(B\) is an standard Brownian motion and \(\sigma\) is a positive process that satisfies a certain Hölder continuity condition. This condition includes in particular the local volatility case \(\sigma_t:=\sigma(t,S_t)\) for \(\sigma(t,x)\) Hölder continuous with respect to \(x.\) The authors search for a sequence of \(N_T\) stopping times \(\tau_i\) on \([0,T],\) with \(N_T<\infty\), a.s., that minimizes the quadratic variation of the hedging error \[ Z_t:=\int_0^t \partial_x u(s,S_s)dS_s-\sum_{\tau_{i-1}\leq t} \partial_x u(\tau_{i-1},S_{\tau_{i-1}}) (S_{t \wedge \tau_i}-S_{\tau_{i-1}}), \] where \(u\) represents the price of a derivate. Recall that the quadratic variation allows one to control the law of \(\sup_{t\in [0,T]} |Z_T|\) thanks to the Lenglart inequality. It is well known that a deterministic regular hedging with \(N_T\) rebalancing times, for payoffs smooth enough, gives a convergence of \({\mathbb E}(\left\langle Z\right\rangle_T)\) to zero at rate \(N^{-1}_T\). This justifies the fact that the authors try to minimize a.s. \(N_T \left\langle Z\right\rangle_T.\) \textit{M. Fukasawa} [Prog. Probab. 65, 331--346 (2011; Zbl 1246.91130)] founded an asymptotic lower bound and an optimal strategy, minimizing \({\mathbb E}(N_T) {\mathbb E}(\left\langle Z\right\rangle_T)\). Here, the authors obtain similar results but minimizing a.s. \(N_T\left\langle Z\right\rangle_T\) and moreover, for a much larger class of payoffs. Concretely the consider payoffs \(g(S_T, Y_T)\) where \(Y_T\) is a functional of the whole trajectory of \(S\). This includes for example, Asian and lookback options. Given a square summable sequence \(\varepsilon_n\), converging to \(0,\) that is assumed to control simultaneously, the closeness of consecutive values of \(S\) on the rebalancing dates and the increasing of \(N_T\), the authors obtain the following a.s. lower bound (Theorem 3.1): \[ \liminf_{n\rightarrow \infty} N_T^n \langle Z^n \rangle_T \geq \left(\int_0^T \frac{1}{\sqrt{6}}\sigma^2(t,S_t)|\partial_{xx} u(t,S_t)| dt\right)^2. \] The result of Fukusawa in [loc. cit.] was \[ \liminf_{n\rightarrow \infty} {\mathbb E}(N_T^n) {\mathbb E}(\langle Z^n \rangle_T) \geq \left({\mathbb E}\left(\int_0^T \frac{1}{\sqrt{6}}\sigma^2(t,S_t)|\partial_{xx} u(t,S_t)| dt\right)\right)^2. \] Furthermore, they obtain the following optimal strategy (Theorem 3.2): \[ \tau_i^n:=\inf\{ t\geq \tau_{i-1}^n: |S_t-S_{\tau^n_{i-1}}|>\root{\varepsilon_n}\of{\frac{|\partial_{xx}u(\tau_{i-1}^n, S_{\tau_{i-1}^n})|}{\sqrt{6}}+\mu \chi(\frac{|\partial_{xx}u(\tau_{i-1}^n, S_{\tau_{i-1}^n})|}{\mu\sqrt{6}})}\}\wedge T, \] where \(\mu\) is a positive constant and \(\chi\) is a smooth function such that \(\chi(x)=1\) for \(x\leq \frac{1}{2}\) and \(\chi(x)=0\) for \(x\geq 1.\) The use of \(\mu\) allows one to avoid the hypothesis \[ {\mathbb P}(\inf_{0\leq t\leq T} |\partial_{xx} u(t,S_t)|>0)=1. \] In the case this hypothesis is true, we can put \(\mu=0\) and the same optimal strategy as in [loc. cit] is obtained. The proofs are based on some lemmas and propositions related with almost sure convergence presented in the paper. The last section of the paper is devoted to numerics, including a practical description of how to build an optimal sequence of stopping times attaining the lower bound. An example of an exchange binary option, not covered by results in [loc. cit], is treated in detail.