Same Book, Different Fills: Partial Identification of FIFO Execution from Aggregate Order Books

Battle Count: 70

Highly relevant to quantitative trading. The paper directly addresses a fundamental problem in execution simulation: L2-based backtests depend on unobservable queue assumptions. It provides a rigorous framework for quantifying how much execution policy conclusions depend on cancellation allocation rules, demonstrates that policy choices can change across FIFO realizations, and offers practical guidance (sensitivity analysis rather than single-point estimates). The findings have direct implications for execution algorithm design, simulator validation, and the decision of whether to invest in MBO data infrastructure.

High complexity. Requires: (1) L1/L2 feed synchronization and trade reconciliation, (2) order-level queue reconstruction with latent partitioning, (3) implementation of three cancellation allocation rules (front, quantity-weighted-random, back), (4) virtual queue marker tracking for passive orders, (5) integration with JAX-LOB GPU matching engine for independent verification, (6) careful handling of boundary conditions, session phases, and unresolved transitions, (7) matched episode construction across rules/seeds/sides, (8) date-resampling uncertainty quantification. The mathematical framework (partial identification, conditional ordering propositions) adds theoretical complexity.

The methodology is well-described with precise mathematical formulations, compiler rules, and experimental protocol. However, the study relies on proprietary LSEG Tick History data for Tokyo Stock Exchange, which is not publicly available. The JAX-LOB matching engine is open-source, but the full reconstruction pipeline and data processing are not released. The chronological protocol (Jan-May estimation, June selection, July held-out) is clearly specified, aiding methodological reproducibility even if exact numerical results cannot be replicated without the data.