Paper: arXiv 2407.13687

Authors: Jing Xu, Yung-Cheng Hsu, William Biscarri

Abstract

Securities lending is an important part of the financial market structure, where agent lenders help long term institutional investors to lend out their securities to short sellers in exchange for a lending fee. Agent lenders within the market seek to optimize revenue by lending out securities at the highest rate possible. Typically, this rate is set by hard-coded business rules or standard supervised machine learning models. These approaches are often difficult to scale and are not adaptive to changing market conditions. Unlike a traditional stock exchange with a centralized limit order book, the securities lending market is organized similarly to an e-commerce marketplace, where agent lenders and borrowers can transact at any agreed price in a bilateral fashion. This similarity suggests that the use of typical methods for addressing dynamic pricing problems in e-commerce could be effective in the securities lending market. We show that existing contextual bandit frameworks can be successfully utilized in the securities lending market. Using offline evaluation on real historical data, we show that the contextual bandit approach can consistently outperform typical approaches by at least 15% in terms of total revenue generated.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper applies existing contextual bandit frameworks (like Thompson Sampling) without extensive novel mathematical derivations, focusing on adaptation to a specific market structure. However, it demonstrates strong empirical rigor by using offline evaluation on real historical data to quantify a consistent 15% revenue improvement over baseline methods.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Optimize Agent Lender Revenue<br>via Dynamic Pricing"] --> B["Data Source<br>Historical Securities Lending Market Data"]
  B --> C{"Methodology Comparison"}
  C --> D["Baseline: Static Rules /<br>Supervised ML Models"]
  C --> E["Proposed: Contextual Bandit<br>for Adaptive Pricing"]
  
  D & E --> F["Offline Evaluation<br>Simulated Trading on Historical Data"]
  
  F --> G["Key Finding<br>Contextual Bandits Outperform"]
  G --> H["+15% Total Revenue<br>Scalable & Adaptive to Market Conditions"]

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 applying contextual bandit frameworks to dynamically price securities lending offers, aiming to improve revenue optimization for agent lenders.

Source sentence
We show that existing contextual bandit frameworks can be successfully utilized in the securities lending market.

Assumes 6

  • The offline evaluation uses historical market supply‑demand data as context, assuming it approximates real‑time conditions. data
    Source sentence
    In this work, we use historical market supply-demand data as context to evaluate multiple contextual bandit strategies.
  • Market conditions are non‑stationary, so offline results may not fully transfer to real‑time applications. market
    Source sentence
    In this work, we use historical market supply-demand data as context to evaluate multiple contextual bandit strategies. Yet, in reality with real-time requests, it’s nearly impossible to obtain such data.
  • A single agent lender’s pricing decisions do not systematically impact aggregate market demand. model
    Source sentence
    In a competitive market, the revenue optimization problem for the representative agent lender in the marketplace takes market demand as given.
  • The reward (preference) function balances hit rate versus ask price to reflect business metrics. model
    Source sentence
    For example, in SL market, the business prefers high acceptance rate of orders flow and higher price conditional on acceptance.
Show 2 more
  • Agent lender market share captures pricing power dynamics (monopoly vs low share). market
    Source sentence
    When an agent lender owns a dominating share of the market, the agent lender has tremendous pricing power, the lending fee tends to be more aggressive than when the agent lender has low market share.
  • Only one agent lender is considered, ignoring potential price wars among multiple lenders. execution
    Source sentence
    We mainly focus on how one agent lender could best setup their pricing strategy internally, where they only consider the available market data along with the prices.

Data

Backtest sample: securities lending market · other · daily

  • NGT autoborrow bid‑level data order-book proprietary
    U.S. securities with lending fee 1%‑10%
    Source sentence
    We trained our models with historical bid-level data from NGT autoborrow, so that each observation contains a timestamp, quantity, bid, and accept or reject status.
  • Additional market feature data alternative proprietary
    Source sentence
    Additional data for market features including market share, utilization, other source of supply, borrower’s bid from the demand side is obtained separately from other sources.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The contextual bandit approach consistently outperforms typical pricing approaches by at least 15% in total revenue. performance
    Source sentence
    the contextual bandit approach can consistently outperform typical approaches by at least 15% in terms of total revenue generated.
  • Contextual bandit‑based methods almost uniformly outperform traditional non‑contextual bandit pricing policies across multiple 5‑day test periods. performance
    Source sentence
    the contextual bandit‑based methods almost uniformly outperform the traditional non‑contextual bandit pricing policies.
  • Policies using the custom reward function with logistic‑regression estimated booking status perform better than those estimating revenue directly. performance
    Source sentence
    the policies which utilized our custom reward function by estimating the booking status using logistic regression performed better than the framework which estimated revenue directly.

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.

Cited by 1 paper in the archive (1 by independent authors)

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.

Cited References (30)

  1. Proposed Updates to Securities Lending Market (2021) — link
  2. J.P. Morgan Asset Management Securities Lending Overview (2023) — link
  3. Over-the-counter market liquidity and securities lending (2019) — link
  4. Securities lending, shorting, and pricing (2022) — link
  5. A contextual-bandit approach to personalized news article recommendation (2010) — link
  6. Offline Deep Reinforcement Learning for Dynamic Pricing of Consumer Credit (2022) — link
  7. Reinforcement learning for optimized trade execution (2006) — link
  8. DeepLOB: Deep Convolutional Neural Networks for Limit Order Books (2019) — link
and 22 more
  1. Market Making under Order Stacking Framework: A Deep Reinforcement Learning Approach (2022) — link
  2. Addressing Non-Stationarity in FX Trading with Online Model Selection of Offline RL Experts (2022) — link
  3. Reinforcement learning applied to Forex trading (2018) — link
  4. Optimal, truthful, and private securities lending (2021) — link
  5. Private Information, Securities Lending, and Asset Prices (2021) — link
  6. Market Power in the Securities Lending Market (2022) — link
  7. Low-Rank Bandit Methods for High-Dimensional Dynamic Pricing (2019) — link
  8. Semi-Parametric Dynamic Contextual Pricing (2019) — link
  9. Dynamic Online Pricing with Incomplete Information Using Multi-Armed Bandit Experiments (2018) — link
  10. Unbiased offline evaluation of contextual-bandit-based news article recommendation algorithms (2011) — link
  11. Sample-efficient Nonstationary Policy Evaluation for Contextual Bandits (2012) — link
  12. Doubly Robust Policy Evaluation and Optimization (2014) — link
  13. Improving offline evaluation of contextual bandit algorithms via bootstrapping techniques (2014) — link
  14. Deciding equivalances among conjunctive aggregate queries (2007) — link
  15. Scilab — link
  16. Scilab — link
  17. The Computational Geometry Algorithms Library — link
  18. The Computational Geometry Algorithms Library — link
  19. The Parmap library — link
  20. The Parmap library — link
  21. BabyJ : from Object Based to Class Based Programming via Types (2003)
  22. tag-sound

0 of 30 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.