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

Download:

Current browse context:

econ.GN

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

Economics > General Economics

Title: Computational Efficiency in Multivariate Adversarial Risk Analysis Models

Abstract: In this paper we address the computational feasibility of the class of decision theoretic models referred to as adversarial risk analyses (ARA). These are models where a decision must be made with consideration for how an intelligent adversary may behave and where the decision-making process of the adversary is unknown, and is elicited by analyzing the adversary's decision problem using priors on his utility function and beliefs. The motivation of this research was to develop a computational algorithm that can be applied across a broad range of ARA models; to the best of our knowledge, no such algorithm currently exists. Using a two-person sequential model, we incrementally increase the size of the model and develop a simulation-based approximation of the true optimum where an exact solution is computationally impractical. In particular, we begin with a relatively large decision space by considering a theoretically continuous space that must be discretized. Then, we incrementally increase the number of strategic objectives which causes the decision space to grow exponentially. The problem is exacerbated by the presence of an intelligent adversary who also must solve an exponentially large decision problem according to some unknown decision-making process. Nevertheless, using a stylized example that can be solved analytically we show that our algorithm not only solves large ARA models quickly but also accurately selects to the true optimal solution. Furthermore, the algorithm is sufficiently general that it can be applied to any ARA model with a large, yet finite, decision space.
Subjects: General Economics (econ.GN); Computation (stat.CO)
Journal reference: Decision Analysis 16.4 (2019): 314-332
Cite as: arXiv:2110.12572 [econ.GN]
  (or arXiv:2110.12572v1 [econ.GN] for this version)

Submission history

From: Michael Perry [view email]
[v1] Mon, 25 Oct 2021 01:24:46 GMT (502kb)

Link back to: arXiv, form interface, contact.