← All workOngoing

Listed derivatives · India · L2 LOB · HMM

Project Tomography

Infer latent queue dynamics and short-horizon flow from noisy Level-2 data, with explicit uncertainty and failure modes.

MicrostructureIndiaHMM

Context

A buys-side desk needed more than descriptive LOB stats: they wanted a repeatable way to flag when the book was consistent with informed flow versus noise, without overfitting to a single session.

Approach

  1. Clean and align tick-level L2 with session rules and auction mechanics documented in the memo.
  2. Specify a compact hidden-state model for quote updates and trade arrivals; compare to simpler baselines.
  3. Report posterior summaries and economic interpretation — not just in-sample fit.

Outcomes

  • Written model specification + estimation note suitable for risk and compliance review.
  • Playbook for when the signal is weak (thin markets, halts, roll windows).

Tools & stack

Python · NumPy/Pandas · Custom HMM estimation · Parquet / tick store