Understanding Dealers and their Counterparty Networks in Over-The-Counter (OTC) Markets

Abstract

Trading in Over-The-Counter (OTC) markets is facilitated by dealers. They play an important role in matching clients, both buyers and sellers, stabilizing prices, and providing liquidity. Traditionally, research on OTC trading has had a focus on understanding aggregate market outcomes. In this paper, we address a complementary task that shifts the focus to the dealers, and the impact of their actions. We apply machine learning methods to model and predict the future trading actions of individual OTC dealers. Using a historical time-series dataset of the daily trading activity for each dealer, over a set of US corporate bonds, we address two key tasks for dealers. The rst is to determine the bonds that the dealer will trade in some future interval. The second is to predict the future counterparty network of each dealer. We consider a range of neural network based prediction models and we enhance them using novel bond-bond and dealer-dealer embeddings. An extensive evaluation highlights modeling successes, and opportunities to improve the models. A mini-case study identifi es a subset of dealers whose behavior patterns may be modeled by strategic agents, thus opening up new avenues for machine learning research in financial markets.