Paper: arXiv 2506.18147

Authors: Paloma Marín, Sergio Ardanza-Trevijano, Javier Sabio

Abstract

The digitalization of financial markets has shifted trading from voice to electronic channels, with Multi-Dealer-to-Client (MD2C) platforms now enabling clients to request quotes (RfQs) for financial instruments like bonds from multiple dealers simultaneously. In this competitive landscape, dealers cannot see each other’s prices, making a rigorous analysis of the negotiation process crucial to ensure their profitability. This article introduces a novel general framework for analyzing the RfQ process using probabilistic graphical models and causal inference. Within this framework, we explore different inferential questions that are relevant for dealers participating in MD2C platforms, such as the computation of optimal prices, estimating potential revenues and the identification of clients that might be interested in trading the dealer’s axes. We then move into analyzing two different approaches for model specification: a generative model built on the work of (Fermanian, Guéant, & Pu, 2017); and discriminative models utilizing machine learning techniques. Our results show that generative models can match the predictive accuracy of leading discriminative algorithms such as LightGBM (ROC-AUC: 0.742 vs. 0.743) while simultaneously enforcing critical business requirements, notably spread monotonicity.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced probabilistic graphical models and causal inference (structural equations, do-calculus analogs) to model RfQ dynamics, justifying high math complexity. Empirical rigor is strong, evidenced by backtest-ready analysis on a proprietary BBVA dataset comparing generative vs. discriminative models (ROC-AUC metrics), though it lacks full production backtesting code or live implementation details.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Analyze RfQ Process & Dealer Profitability"] --> B["Data Input<br>MD2C Platform Trading Logs"]
  B --> C["Methodology<br>Probabilistic Graphical Model"]
  C --> D{"Computational Approaches"}
  D --> E["Generative Model<br>Spread Monotonicity Constraints"]
  D --> F["Discriminative Model<br>LightGBM ML Algorithm"]
  E & F --> G["Outcomes<br>Comparative Analysis"]
  G --> H{"Key Findings"}
  H --> I["Generative & Discriminative<br>Accuracy Comparable"]
  H --> J["Generative Model Enforces<br>Spread Monotonicity"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a unified framework based on probabilistic graphical models and causal inference for analyzing RfQ processes, including a generative model that matches leading discriminative algorithms while enforcing spread monotonicity.

Source sentence
This article introduces a novel general framework for analyzing the RfQ process using probabilistic graphical models and causal inference.

Assumes 6

  • Price monotonicity is enforced in the generative model, meaning clients prefer more favorable quotes. model
    Source sentence
    approach naturally enforces economically consistent behaviors—such as price monotonicity, whereby clients prefer more favorable quotes—and seamlessly integrates…
  • Volatility is treated as a relatively predictable quantity and pre‑trade and post‑trade volatility are effectively the same. market
    Source sentence
    we treat volatility as a relatively predictable quantity [17]; therefore, we assume that pre‑trade and post‑trade volatility are effectively the same variable.
  • Clients have no information asymmetry (IA = 0) for benchmarking purposes. data
    Source sentence
    We assume that clients in the dataset do not possess information asymmetry, i.e., we set IA = 0.
  • All passed RfQs are treated as trade‑intended (PD = 0) to simplify modeling. data
    Source sentence
    For simplicity, we assume that price discovery is not the motivation in the passed RfQs, but they correspond to situations where none of the quoted prices
Show 2 more
  • RfQ arrivals follow a Poisson process with intensity λ. model
    Source sentence
    As it is usual in the optimal market‑making literature [28,29], we model the arrival of RfQs as a Poisson process Nt with intensity λ.
  • A minimal conditioning set Zt satisfies the back‑door criterion, blocking spurious correlations. statistical
    Source sentence
    The choice of features is critical to ensure that the results are not driven by spurious correlations. We guide our feature selection based on the findings from previous sections regarding the minimal conditioning set that satisfies the back‑door criterion.

Data

  • BBVA RfQ dataset · BBVA other proprietary
    Source sentence
    We use a proprietary dataset of RfQs from BBVA, consisting of 102,437 RfQs, of which 5,738 are labeled as hits.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The generative model satisfies the business constraint of spread monotonicity, unlike the discriminative models. other
    Source sentence
    it uniquely satisfies the key business constraint of spread monotonicity—an essential property for applications such as optimal pricing.
  • Both generative and LightGBM models substantially outperform the majority‑class baseline (AUC‑ROC 0.5). predictive-accuracy
    Source sentence
    logistic regression. While the generative model exhibits lower calibration performance
  • The ROC curves show the generative model performs comparably to LightGBM and both outperform logistic regression. predictive-accuracy
    Source sentence
    the generative model performs comparably to LightGBM, and both outperform logistic regression.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 12 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.