A theoretical framework for the pricing of contingent claims in the presence of model uncertainty (Q997952): Difference between revisions

The authors generalize the classical Black-Scholes finance model, in order to handle the pricing of European contigent claims in a finance market controlled by a family of fair gambling strategies. This setting is a generalization of the uncertain volatility model (UVM in abbrev.) introduced by \textit{M. Avellaneda, A. Levy} and \textit{A. Paras} [Pricing and hedging derivative securities in markets with uncertain volatilities. Appl. Math. Finance 2, 73--88 (1995)]. The underlying model of uncertainty is modelled by a family of ``martingale measures'' given on a measurable space \((\Omega,{\mathcal B})\), where \(\Omega(=C_0[0,T])\) denotes the Banach space of real functions (null at ``\(t=0\)'') defined on the compact interval \([0,T]\) and then equipped with its supremum norm. Let \({\mathcal B}\) denote the Borel \(\sigma\)-field on \(\Omega\), by martingale measure we mean a probability law \(P\) on \((\Omega,{\mathcal B})\) such that the coordinate process, \[ \begin{cases} B=(\Omega,({\mathcal B}_t)_{t\in[0,T]},B(\omega,s)= \omega(s),P),\text{ where }\omega\in\Omega,\;s\in [0,T],\\ \text{and }({\mathcal B}_t)_{t\in[0,T]}\text{ denotes for the canonical filtration in } (\Omega,{\mathcal B})\end{cases},\tag{1} \] turns out to be a martingale. Let \({\mathcal P}_m\) denote the class of martingale measures on \((\Omega, {\mathcal B})\) and denote by \({\mathcal P}\) the subclass of members of \({\mathcal P}_m\) which verify the following property, \[ \begin{cases}\text{There exists a continuous distribution }\mu(ds)\text{ on }[0,T]\\ \text{such that the process }(\langle B\rangle^P-\underline\mu)\text{ (resp., }\overline\mu-\langle B\rangle^P)\\ \text{is increasing for each }P\in{\mathcal P}, \text{ in which }\langle B \rangle^P\text{ denotes}\\ \text{the quadratic variation of the coordinate process }B.\end{cases}\tag \(\text{H}(\underline\mu)\) \] then it is possible to associate here a Choquet capacity on \((\Omega, {\mathcal B})\) by putting, \[ c(f):=\sup\{\|f \|_{L^2(P)}\mid P\in{\mathcal P}\|,\;\forall f\in{\mathcal B}\text{-measurable and bounded}.\tag{2} \] A subset \(A \subset\Omega\) is called polar provided that \(A\) is contained in a set \(B \in{\mathcal B}\) such that \(c(B)=0\). A mapping \(f\) on \(\Omega\) with value in a topological space is called quasi-continuous provided that, \[ \forall \varepsilon>0,\;\exists{\mathcal O}\subset\Omega\text{ open such that }c( {\mathcal O})<\varepsilon\text{ and }f\text{ is continuous on }(\Omega \setminus {\mathcal O}). \tag{3} \] Stochastic integrals w.r.t. \({\mathcal P}\). Since the family of martingale measures \({\mathcal P}\) may be unbounded, in order to extend the Ito-isometry from the class of ``predictable step-processes over \(({\mathcal B}_t)_{t \in[0,T]}\)'' to the class of ``predictable stochastic integrands'' the authors assume furthermore the following hypothesis \(H (\overline\mu)\), \[ \begin{cases}\text{There exists a continuous distribution }\overline{(\mu)}(ds)\text{ on }[0,T]\\ \text{s.t. the process }(\overline \mu-\langle B\rangle^P)\text{ is increasing for each }P\in {\mathcal P},\\ \text{in which }(B)^P\text{ denotes the quadratic variation of }B.\end{cases}\tag{4} \] Then by taking account of the capacity \(c (\cdot)\) given in (2) above, they present a method of construction of Ito's isometry similar to that of the classical case as follows. Theorem 2.8 (in the paper under review). Assume \(H(\underline\mu)\) and \(H(\overline\mu)\). Let \(h\) be a step-process of the form \(h= \sum^N_{i=1}k_{t_i} \chi_{]t_i,t_{i+1}]}\) and define, \[ I_T(h):= \sum^N_{i=1}k_{t_i}(B_{t_{i+1}}-B_{t_i})\left(=\int^T_0h_s\,dB_s \right),\tag{5a} \] to be the stochastic integral of \(h\). a) Then \(I_T\) defines a linear mapping on the space \({\mathcal H}_e\) of step-processes which could be extend to be a linear mapping from the space \({\mathcal H}\) of predictable integrands into the Banach space \({\mathcal L}\) of quasi-continuous and quasi-bounded functions on \(\Omega\). b) \({\mathcal H}\) could be obtained as the completion of \({\mathcal H}_e\) w.r.t. the following norm, \[ \|h\|^2_{\mathcal H}=\sup \left\{E_P\left(\int^T_0h^2_s.ds\right) \mid P\in\wp \right\}\tag{5b} \] and furthermore, \[ c(I_T(h))\leq \|h\|_{\mathcal H},\;\forall\in {\mathcal H}.\tag{5c} \] Cheapest riskless superreplication price. By convention we do not distinguish in the Banach space \({\mathcal L}\) functions which are defined quasi-everywhere with their equivalence classes. In their financial context the authors interpret each member \(f\in{\mathcal L}\) as a contingent claim, i.e., a cheapest riskless superreplication price that we are interested in. Put, \[ \Lambda(f)=\inf\{a\in\mathbb R \mid\exists h\in{\mathcal H}\text{ such that }a+I_T(h)\geq f\text{ quasi-everywhere}\},\tag{6a} \] the main results of the paper is to characterize out those members of, \[ \text{dom} (\Lambda):=\{f\in{\mathcal L}\mid \Lambda(f)<+\infty\}, \tag{6b} \] which verify the following ``duality principle'' is, \[ \begin{cases}\Lambda(f)= \inf\{E_P (f)\mid f\in\wp'\}\\ \text{where }\wp'\text{ is a given subset of }\wp \end{cases}\tag{6c} \] Main Results. Theorem 3.1. a) Assume that \(H (\underline \mu)\) and \(H(\overline\mu)\) are verified, then there are subclasses \[ {\mathcal P}'' \subset{\mathcal P}_m\text{ and }{\mathcal K}\subset C_b (\Omega),\tag{7a} \] such that, \[ \Lambda(f)=\inf\{E_P(f)\mid P\in \wp''\}\text{ for each }f\text{ given in }{\mathcal K}.\tag{7b} \] b) In the particular case where \(\text{dom}(\Lambda)={\mathcal L}\), then \[ \begin{cases} \Lambda(f)=\inf\{E_P(f)\mid f\in \wp_{pol}\}\text{ for }\forall f\in{\mathcal L},\text{ where }\wp_{pol}(\subset \wp_m)\\ \text{denotes the class of martingale measures which do not charge polar sets.}\end{cases}\tag{7c} \] Theorem 6.1. Assume that \(H(\underline\mu)\) and \(H(\overline\mu)\), furthermore \(d\underline\mu\leq d\overline\mu\), and \(\overline \mu(ds)\) is Hölder continuous. Then, \[ \begin{cases}\Lambda(f)=\inf\{E_P(f)\mid P\in \wp\}\text{ for each }f\text{ given in }{\mathcal K}\cap{\mathcal C}_b(\Omega), \text{ in which}\\ \wp\text{ denotes the class of member of }\wp\text{ which verify }H (\underline\mu)\text{ and }H(\overline\mu).\end{cases} \tag{8} \] From the conclusion of the authors: ``In the case of the UVM, this characterization is complete and it works for a large class of European path-dependent claims. This characterization in the general setting remains an open question''. Note that the UVM corresponds precisely to the case where the measures \(\underline\mu\) and \(\overline\mu\) are both identified with the Lebesgue measure on \([0,T]\). Reviewer's remark: In the proof of Theorem 3.1 (see above) the authors assume firstly that \(\Omega\) is bounded and closed w.r.t. its supremum norm. This seems to be not the case, since even in the UMV case we are already concerned with the trajectories of recurrent Brownian motions. The reviewer thinks that it be would more reasonable to consider a subspace \(\Omega_T(\subset\Omega)\) which consists of ``trajectories stopped at the exit-time \(T\) of open set of \(\mathbb{R}\)'', in such a way that the capacity of an\((\Omega\subset\Omega_T)\) is arbitrary small.