Optimization heuristic for large-scale B2B integral debt netting

Deferred payments are widely used in the business-to-business (B2B) economy as an important short-term financing source. However, the resulting pending invoices create complex debt webs and may induce cascading liquidity shortages that jeopardize the system's stability. This paper presents a framework for integral debt netting (that is, without fragmenting debts) that aims to lower liquidity needs by settling selected invoices thanks to an external funder. This framework models B2B obligations as weighted directed multi-graphs and uses various metrics to assess how effectively external financing enables invoice clearing. In addition, we propose several extensions to provide scalability to large problem instances as well as adaptability to constraints over the working capital. We describe a specific division strategy to manage large datasets while preserving the results' quality and providing limited computation times. A set of experiments with synthetic and large-size real data allows us to evaluate the behavior of our process and the quality of its results. They show that our algorithm obtains a good trade-off between the netting quality and the network's coverage.