Extend causal inference to the multi-RfQ optimization setting

Develop a rigorous causal inference framework for the multi-RfQ optimization problem in Multi-Dealer-to-Client bond trading platforms, where revenues are linked across multiple quotes rather than isolated to single events, by formulating identifiable interventional quantities (such as hit probabilities and revenues on hit) under appropriate conditioning and specifying estimation procedures from observational data.

Background

In the single-RfQ setting, the paper identifies minimal conditioning sets that enable unbiased estimation of causal effects (e.g., the hit probability) from observational data. However, in practice dealers face multi-RfQ dynamics, where inventory risk and future flow link revenues across multiple quotes, complicating causal identification.

The authors note that standard market-making formulations often adopt simplified hit probability structures, and that ignoring confounding can bias optimal pricing decisions. They argue that extending the causal framework to the multi-RfQ context is necessary to properly analyze interventions when quotes are dynamically interdependent.