Strategic Learning and Trading in Broker-Mediated Markets

Alif Aqsha, Fayçal Drissi, Leandro Sánchez-Betancourt

TL;DR

Brokers hold a massive strategic advantage because they see their clients' "cards"—specifically, the order flow of informed traders. This paper models how brokers use filtering techniques to turn this information leakage into a speculative edge, finding that the economic value of seeing client flow is comparable in magnitude to the transaction fees they charge.

What Problem Does This Paper Address?

In financial markets, information is the ultimate currency. Usually, we think of "informed traders" (like hedge funds) having the upper hand because they possess private signals about where a price is headed. However, these traders often have to go through a broker to execute their trades.

The broker sits in a unique position: they provide liquidity to the informed trader (acting as the counterparty) while simultaneously managing their own inventory in the public ("lit") market. This creates a strategic "cat-and-mouse" game. The trader tries to profit from his signal without revealing too much, while the broker tries to "filter" the trader's activity to steal that signal for herself.

The authors address a critical gap: How do these two agents optimally learn from and exploit each other in a dynamic, partial-information environment? Specifically, they quantify the advantage a broker gains by observing client flow versus just looking at market prices.

Methods

The authors employ a sophisticated mathematical framework combining Stochastic Control and Filtering Theory:

  • Kalman-Bucy Filtering: Both agents use this to estimate unobserved variables. The informed trader tries to estimate the broker's public market activity from noisy price changes. The broker tries to estimate the trader's private signal, either from market prices or from the trader's direct order flow.
  • Hamilton-Jacobi-Bellman (HJB) Equations: To find optimal strategies, the authors solve HJB equations. Because the model uses a linear-quadratic structure, these simplify into systems of Riccati Differential Equations (RDEs), which allow for explicit, closed-form solutions.
  • Stackelberg-type Formulation: The model assumes a hierarchy where the broker anticipates the informed trader's optimal response, allowing her to strategically influence the trader's beliefs.
  • Internalization vs. Externalization: The broker must decide whether to keep a client's trade on her own books (internalize) or immediately hedge it in the public market (externalize).

Datasets

The paper does not use historical market data. Instead, it relies on extensive numerical simulations (10,000 runs) to validate the theoretical models. The parameters for these simulations (volatility, price impact, mean reversion) are calibrated based on established literature in algorithmic trading (e.g., Cartea and Sánchez-Betancourt, 2025).

Key Insights and Ideas

  • Prices are Noisy, Flow is Clear: Market prices have a low signal-to-noise ratio. If a broker only relies on prices to guess what an informed trader knows, her performance is no better than a "naive" strategy. However, client order flow is a "high-fidelity" signal that allows the broker to reconstruct the private signal almost perfectly.
  • The "Information Tax": The value the broker extracts from "learning" the client's signal is roughly equal to the transaction costs. For the informed trader, this is a hidden cost of doing business with a single broker.
  • Strategic Manipulation: In a "second-order" analysis, the authors show the broker can intentionally trade in the public market to mislead the informed trader. By creating "fake" price movements, the broker can trick the trader into trading in the opposite direction of her own inventory needs.
  • Privacy through Distribution: The results explain why sophisticated traders often split their orders across multiple brokers (broker aggregation)—it is a necessary defense to prevent any single broker from "filtering" their private signal.

Key Results

  • Outperformance: When the broker filters information from client flow, she significantly outperforms all benchmarks (e.g., externalizing everything or simple TWAP unwinding).
  • Externalization Rates: Brokers typically externalize (hedge) 70% to 90% of informed flow. They are more likely to "trust" the signal and trade aggressively when the signal's mean-reversion is high (i.e., the signal is less likely to diffuse away).
  • Robustness: The broker's advantage remains stable even if she slightly mis-specifies the model parameters (like the signal's volatility), suggesting that the "flow-based" edge is highly resilient.

Limitations and Future Work

  • Single-Broker Assumption: The current model assumes the informed trader uses only one broker. In reality, traders use many, which dilutes the signal each broker sees.
  • Asymmetric Sophistication: The model assumes the informed trader is "first-order" strategic (he doesn't fully account for the fact that the broker is learning from him). A more complex model would involve a full Nash equilibrium of mutual learning.
  • Future Directions: The authors suggest exploring an "equilibrium of beliefs," where both players adjust their internal models of each other over time until their expectations match the realized market dynamics.

Summary updated

Abstract

We study strategic interactions in a broker-mediated market in which agents learn and exploit each other's private information. A broker provides liquidity to an informed trader and to noise traders while managing inventory in a lit market. The informed trader infers the broker's trading activity in the lit market, while the broker estimates the trader's private signal. Information leakage in the client's trading flow generates economic value for the broker that is comparable in magnitude to transaction costs: the broker can speculate profitably and manage risk more effectively, which in turn adversely affects the informed trader's performance. Brokers therefore hold a strategic advantage over traders who rely solely on prices to filter information. When the broker only relies on prices rather than client trading flow to infer information, their trading performance becomes indistinguishable from the performance of a naive strategy that internalises noise flow, externalises informed flow, and offloads inventory at a constant rate.
BibTeX
@misc{aqsha2024strategic,
  author = {Alif Aqsha and Fayçal Drissi and Leandro Sánchez-Betancourt},
  title = {{Strategic Learning and Trading in Broker-Mediated Markets}},
  year = {2024},
  eprint = {2412.20847},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2412.20847},
}
Figure