Dice Question Streamline Icon: https://streamlinehq.com

Embed cover price information into discriminative models

Develop effective feature engineering or modeling techniques to incorporate post-trade cover price information into discriminative models (such as logistic regression or gradient-boosted trees) for predicting hit probabilities in MD2C RFQ data, and evaluate their impact on calibration and predictive performance.

Information Square Streamline Icon: https://streamlinehq.com

Background

The authors experimented with engineered features derived from cover prices, such as rolling averages of the distance between the dealer’s quote and the observed cover price, but found no significant predictive improvement.

Given the importance of post-trade signals (e.g., cover prices) in MD2C negotiations, the paper highlights the need for better approaches to integrate such information into discriminative models without violating causal or economic constraints.

References

Developing more effective approaches to embed cover price information into discriminative models through feature engineering remains an open area of research.

— Causal Interventions in Bond Multi-Dealer-to-Client Platforms (2506.18147 - Marín et al., 22 Jun 2025) in Section "Generative versus discriminative models for causal interventions: empirical results" (feature selection discussion)