Research prototype2025—ongoing

Interpretable market-impact dynamics

A transparent experimental pipeline for isolating the price impact of large orders and discovering compact differential equations that describe it.

Illustrative market impact chart comparing baseline, execution and post-trade predictions

Scope

  • Interpretable ML
  • Financial markets
  • SINDy
  • Simulation
  • Python
View repository ↗

The challenge

Large institutional orders are split into smaller trades. Their price impact accumulates during execution and relaxes afterwards, but ordinary market drift makes that effect difficult to isolate and black-box forecasts are hard to interpret.

The approach

  1. 01Designed paired baseline and treatment simulations from the same initial order-book state and stochastic seed.
  2. 02Defined signed counterfactual impact from the difference between baseline and TWAP trajectories.
  3. 03Built separate controlled sparse-dynamics models for execution and post-trade relaxation.
  4. 04Created explicit data, evaluation and provenance contracts so future empirical claims remain auditable.

The outcome

The research prototype implements the intended pipeline, sparse-regression components and deterministic presentation layer. Its current visual dataset is deliberately illustrative; empirical conclusions await completed and audited simulation runs.

Next case studyBowling detection