We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

q-fin.CP

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantitative Finance > Computational Finance

Title: Pricing high-dimensional Bermudan options with hierarchical tensor formats

Abstract: An efficient compression technique based on hierarchical tensors for popular option pricing methods is presented. It is shown that the "curse of dimensionality" can be alleviated for the computation of Bermudan option prices with the Monte Carlo least-squares approach as well as the dual martingale method, both using high-dimensional tensorized polynomial expansions. This discretization allows for a simple and computationally cheap evaluation of conditional expectations. Complexity estimates are provided as well as a description of the optimization procedures in the tensor train format. Numerical experiments illustrate the favourable accuracy of the proposed methods. The dynamical programming method yields results comparable to recent Neural Network based methods.
Comments: 26 pages, 3 figures, 5 tables, added affiliations and update acknowledgements
Subjects: Computational Finance (q-fin.CP); Computational Engineering, Finance, and Science (cs.CE); Numerical Analysis (math.NA)
MSC classes: 91G60 (Primary) 62J02, 15A69 (Secondary)
Cite as: arXiv:2103.01934 [q-fin.CP]
  (or arXiv:2103.01934v2 [q-fin.CP] for this version)

Submission history

From: Philipp Trunschke [view email]
[v1] Tue, 2 Mar 2021 18:46:28 GMT (42kb,D)
[v2] Sat, 6 Mar 2021 09:19:51 GMT (42kb,D)

Link back to: arXiv, form interface, contact.