R.02 · Financial AI · market microstructure

When Information Acquisition Moves the Market

An inquiry can improve information while changing the execution problem in which that information will be used.

Evidence / ArtifactLink pendingCode / ReproduceLink pending
Status
Research manuscript · 2026
Area
Financial AI · market microstructure
Evidence
  • formal boundary
  • exact enumeration
  • 5,000-world holdout
Method
Exact finite-state experiments and fresh holdout
Fig. 2 — Inquiry as an intervention in RFQ executionSchematic
Schematic: passive versus endogenous value of information in RFQ executionLeft: a trader requests quotes from four dealers and quotes stay fixed. Right: the inquiry changes dealer quotes and the best route switches to a different dealer.PASSIVE VALUE OF INFORMATIONdownstream execution held fixedTRADERD1D2D3D4quotes held fixed while information is valuedQUOTED COSTENDOGENOUS VALUE OF INFORMATIONinquiry changes downstream action valuesTRADERD1D2D3D4route changes → passive and endogenous are not equivalentQUOTED COST
SchematicIn dealer markets, requesting another quote can reveal intent or alter dealer response. A routing change is a direct policy-level certificate that the passive and endogenous formulations are not equivalent. Illustrative, not experimental data.
01 · Question

When does passive valuation choose wrong?

A passive value-of-information calculation holds downstream execution fixed. In dealer markets, requesting another quote can reveal intent or alter dealer response. The work asks when passive information valuation therefore chooses the wrong inquiry or routing policy.

02 · Core idea · formal boundary

Sensing versus acquisition.

The analysis distinguishes an observation-only scalar sensing adjustment from an acquisition action that changes downstream action values. Exact scalar representation is possible only when the inquiry-induced perturbation is constant across downstream actions for a realized observation. A routing change is a direct policy-level certificate that the passive and endogenous formulations are not equivalent.

03 · Selected results

Exact enumeration, fresh holdout.

Heterogeneous targeted-RFQ tests
10,080

heterogeneous targeted-RFQ environments

Fresh IID holdout
5,000

fresh IID holdout environments

Fresh holdout · inquiry breadth
83.2%

inquiry-breadth mismatch on the fresh holdout

Common-response mechanism control
0.0%

routing mismatch under the common-response mechanism control

04 · Experimental design

Exact experiments.

Finite decision problems are solved by exact enumeration. The empirical component uses heterogeneous parameterized environments, exact 3–5 dealer extensions, a fresh 5,000-environment IID holdout, and predefined distribution shifts without changing the solver.

Exact enumeration; separated discovery/holdout seeds; frozen holdout ranges.

05 · Strongest control

Mechanism diagnostics.

Making the inquiry response common across dealers removes routing mismatch. Restricting response intensity to a weak-response regime materially reduces breadth mismatch, while larger orders or shallower liquidity amplify it. This directionality is more informative than a raw average gap.

06 · What the evidence covers

Scope of the evidence.

Formal representability boundary; 10,080 heterogeneous tests; fresh 5,000-world IID holdout; mechanism controls and shifts.

07 · What it does not establish

Not a frequency estimate.

Boundary

The percentages are not estimates of real-market RFQ frequencies. The model is intentionally stylized and does not solve a full strategic multi-dealer equilibrium or calibrate to proprietary transaction data.

08 · Evidence & artifact status

Artifact status.

Reproducibility

Exact enumeration; separated discovery/holdout seeds; frozen holdout ranges.

Boundary

Stylized mechanism test, not an empirical estimate of real RFQ leakage frequency.

Disclosure

Public artifacts can expose the exact mechanism test without exposing the private AEI architecture.

Evidence / ArtifactLink pendingCode / ReproduceLink pending