The Optimizer’s Guide to Money Laundering: Navigating Detection Algorithms with Naïve Beam Search

Abstract

In this paper, our goal is to optimize the total amount of monetary value we can launder within given constraints. We cast the problem into an undirected graph, with various financial institutions represented as nodes. To this end, we assume that the problem is a complete graph and the more we move or “layer” the money, the harder it is for law enforcement to trace back to its illegal origin, reducing risk of detection. As we progress, we also include other prominent components of the money laundering process, such as shell companies. Because each transfer provides different level of layering, for instance, international is harder to trace than national, the expected payoff of each route is a function of total layers minus total cost of moving money. The paper focuses on exploring the rich algorithms and constraints of the money laundering process against detection algorithms. To this end, we review extensively state-of-the-art techniques used by money launderers and state-of-the-art detection algorithms used by financial institutions. We progressively generate more complicated models while retaining as much realism as possible. We first solve the optimization problem for the single best route, assuming that each node can only be travelled through once. We then compare the practicality and calculation time of each method against the CVX calculated upper bound and find that the greedy algorithm family performs exceptionally well. After which, we expand the question to include multiple routes without using each node more than once or less than a given cap when shell companies are incorporated. Naive beam search algorithms prove to be the best performing across all data. Adding a time constraint, we conclude that naive beam search performs over 98.7% of the upper bound and is ideal for path selection due to its low computational complexity.