Monte Carlo methods via a dual approach for some discrete time stochastic control problems

DOI10.1007/S00186-014-0488-3MaRDI QIDQ2264108zbMATH OpenOpenAlexFDO

Authors Lajos Gyurkó, J. H. Witte, Ben M. Hambly

Publication date 20 March 2015

Published in Mathematical Methods of Operations Research (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1112.4351

dual formulation least squares regression Monte Carlo methods stochastic control

Stopping times; optimal stopping problems; gambling theory (60G40) Martingales with discrete parameter (60G42) Applications of Markov chains and discrete-time Markov processes on general state spaces (social mobility, learning theory, industrial processes, etc.) (60J20) Monte Carlo methods (65C05) Derivative securities (option pricing, hedging, etc.) (91G20) Numerical methods (including Monte Carlo methods) (91G60) Financial applications of other theories (91G80) Optimal stochastic control (93E20)

Abstract: We consider a class of discrete time stochastic control problems motivated by some financial applications. We use a pathwise stochastic control approach to provide a dual formulation of the problem. This enables us to develop a numerical technique for obtaining an estimate of the value function which improves on purely regression based methods. We demonstrate the competitiveness of the method on the example of a gas storage valuation problem.

Review text

The numerical pricing of options with early exercise features, such as American options, is a challenging problem, especially when the dimension of the underlying asset increases. In the present paper a class of discrete time stochastic control problems motivated by a range of financial applications is considered. Also, a relatively simple dual approach and which leads to numerical algorithms for the efficient computation of the value function, is developed. A numerical technique based on the dual formulation of the problems to obtain an estimate of the value function is developed. The competitiveness of the method on the example of a gas storage valuation problem is demonstrated. The numerical example shows that the dual formulation based on an upper bound is sharped if we get from the a priori estimate at comparable computational expense. In Section 2 the concept of the standard stochastic control problem is developed. Here, the economy in discrete time defined up to a finite time horizon \(T\) is considered. The financial market is described by the filtered probability space \((\Omega, {\mathcal F}, ({\mathcal F}_t)_{t \in {\mathcal T}}, P),\) where \({\mathcal T} = \{0,1, \dots, T\}.\) Here, \(({\mathcal F}_t)_{t \in {\mathcal T}}\) is the filtration which is generalized by \(X\) and \(R\) is a risk neutral pricing measure. An \({\mathbf R}^d\)-valued discrete time Markov chain representing the \((X_t)_{t \in {\mathcal T}}\) of the underlying assets is taken. The concepts of the \(K-\)admissible exercise policy and the value function are given. In order to indicate the type of problems that fit into this framework for examples are given a Bermudan option, a gas storage valuation, a swing option pricing and an optimal liquidation. In Section 3 the dual problem of the problem to represent the value of the option as the supremum over the set of admissible exercise is presented as a problem for the option value as an infimum over a space of martingal-valued functions. In Theorem 3.1 the representation of the value \(V_0^*(y, X_0)\) of the option is shown. In Lemma 3.3 a version which can be used for the Monte Carlo based numerical technique is derived. In Section 4 numerical implementations of the dual upper bound are realised. In Subsection 4.1 a numerical method that implements the upper estimate derived in Lemma 3.3 is introduced. This is the contents of Algorithm 4.1. Two possible versions of Algorithm 4.1 are presented in Subsections 4.1.1 and 4.1.2. In Subsection 4.2 the concept of a dynamic programming formulation is presented. For the computation of the conditional expectation in the formulation of the dynamic programming it is presented a special algorithm -- Algorithm 4.6. In Subsection 4.2.1 a particular version of Algorithm 4.6 is implemented for the a priori estimate. By Algorithm 4.9 is given a technique to estimate a low biased estimate of \(V_0^*(y_0, \cdot).\) In Section 5 the developed algorithms are practically realised by examples. In Subsection 5.1 the gas storage problem is considered. In Subsection 5.2 the a priori estimate is computed. In Subsection 5.3 the version of the method based on the dual formulation is implemented.

Recommendations

Cites work

Cited in

(12)

This page was built for publication: Monte Carlo methods via a dual approach for some discrete time stochastic control problems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2264108)