Back to past content

Daily Feed - 2026-02-21

Date:

A unified theory of order flow, market impact, and volatility

Domain: Quant Finance / Market Microstructure / Stochastic Processes | Time cost: ~45min read

Intuition: This paper proposes one microstructural mechanism that jointly explains four empirical stylized facts: persistent signed order flow, rough volume, rough volatility, and power-law impact. The key move is decomposing activity into core orders plus reaction flow (both Hawkes processes), then taking a scaling limit.

Concrete punch: One persistence parameter governs all exponents in the limit:

  • signed flow has Hurst index ,
  • traded volume has Hurst index ,
  • volatility has Hurst index ,
  • impact exponent is . Empirically they estimate , which implies impact exponent , i.e., square-root impact.

Significance: Instead of fitting volatility, impact, and order flow with separate ad hoc models, you can calibrate one core persistence statistic and propagate constraints through no-arbitrage structure. That is a practical reduction in model degrees of freedom for execution and risk systems.

Why it matches: Strong physics-lens fit: one latent exponent organizing multiple observables is exactly the kind of universality-style compression you prefer. It is mechanism-first (not benchmark-first) and directly useful for market microstructure modeling.

Author talk search: No direct author talk found yet (searched title + YouTube/Google).


Flow Matching from Viewpoint of Proximal Operators

Domain: ML / Generative Modeling / Optimal Transport | Time cost: ~40min read

Intuition: The paper reframes Optimal-Transport Conditional Flow Matching (OT-CFM) as a proximal-point geometry problem. Instead of viewing the vector field as just a learned transport velocity, it shows the terminal recovery map is exactly a proximal operator of an extended Brenier potential.

Concrete punch: The proximal map appears explicitly:

They show OT-CFM admits an exact formulation of this type (without requiring target density), and prove terminal normal hyperbolicity for manifold-supported targets: after time rescaling, dynamics contract exponentially in directions normal to the data manifold while remaining neutral along tangential directions.

Significance: This gives a rigorous geometric reason why flow-matching trajectories can remain stable near low-dimensional data manifolds. It suggests concrete regularization and solver choices in practice (favoring geometry-preserving dynamics over purely empirical tuning).

Why it matches: Direct hit on your variational/duality taste (Brenier, convex analysis, proximal structure) and the discrete-to-continuous bridge in generative modeling.

Author talk search: No direct author talk found yet (searched title + YouTube/Google).


Liquidation Dynamics in DeFi and the Role of Transaction Fees

Domain: Blockchain / Quant Finance / Mechanism Design | Time cost: ~35min read

Intuition: Liquidations in lending protocols are modeled as a dynamic program where the liquidator can manipulate a constant-product market maker (CPMM) oracle via sandwich-like trades (Oracle Extractable Value, OEV). The paper asks when fee design itself can neutralize that manipulation incentive.

Concrete punch: The liquidator’s control is an intertemporal optimization over trigger/manipulation actions under CPMM execution, with closed-form liquidation bounds. The main claim: there is a fee regime where expected manipulation profit is pushed non-positive (fees are not only a tax, but a security control variable).

Significance: This turns “fee tuning” into a formal security-hardening knob for protocol design, reducing reliance on slower oracle defenses (for example, only time-weighted averaging). It is immediately relevant to protocol parameterization and on-chain risk governance.

Why it matches: Excellent finance↔mechanism-design crossover with explicit dynamic programming structure and practical design implications under adversarial market microstructure.

Author talk search: No direct author talk found yet (searched title + YouTube/Google).


HKML S3E11 — Hawkes Process and Market Microstructure: Too fast but not even furious

Domain: Quant Finance / Point Processes | Time cost: 34min watch

Intuition: Compact seminar-style exposition of how self-exciting point processes explain clustered event arrivals in limit-order-book data and why continuous-price abstractions miss microstructural time effects.

Concrete punch: Uses the standard Hawkes intensity form

with branching ratio as stability criterion. This directly links excitation strength to event clustering and endogenous liquidity shocks.

Significance: Useful as a fast refresher before implementing/calibrating Hawkes-based execution or impact models; aligns with today’s first paper on unified microstructure scaling.

Why it matches: High signal density, mathematically explicit, and directly operational for your market-microstructure stack.


Flash Boys 2.0 — Ari Juels (MIT DCI)

Domain: Blockchain / Market Design / MEV | Time cost: 32min watch

Intuition: A crisp talk framing transaction ordering as a strategic game and explaining why decentralized exchanges can inherit high-frequency-style frontrunning pathologies.

Concrete punch: The key mechanism is ordering optionality: a searcher can insert/reorder transactions if expected extracted value exceeds ordering cost, i.e., exploit when

This is the conceptual bridge to modern MEV/OEV liquidation games.

Significance: Strong conceptual companion to today’s DeFi liquidation paper: it clarifies why oracle manipulation and sandwich-style liquidation strategies arise in the first place.

Why it matches: Good systems+theory blend, mechanism-first framing, and direct relevance to blockchain microstructure research.


Notes on sourcing and recency

  • Papers selected are all from the last month (Jan–Feb 2026).
  • Video picks are older but chosen for pedagogical clarity and direct conceptual linkage to the selected papers.
  • No direct author talks were found yet for today’s three new papers.

Comments