Market Microstructure: The LMF Model & the Square-Root Law of Price Impact

Based on: Sato & Kanazawa (2025), "Why do financial prices exhibit Brownian motion despite predictable order flow?", arXiv:2502.17906v4 [q-fin.TR], Kyoto University.

1. The Central Paradox

In the econophysics of market microstructure, two empirical laws coexist that appear mutually contradictory:

  1. Predictable Order Flow: The market-order flow has long memory — the autocorrelation function (ACF) of order signs decays very slowly:

    • E[εₜ · εₜ₊τ] ∝ τ^−γ, with 0 < γ < 1
    • This means: if many buy orders just arrived, it is likely the next orders will also be buys. This effect persists for hours to days.

  2. Brownian Price Dynamics: Despite this predictable order flow, price movements behave like random Brownian motion — unpredictable, diffusive, consistent with the Efficient Market Hypothesis (EMH).

The question: How can prices be unpredictable if the orders driving them are predictable?

2. The Lillo-Mike-Farmer (LMF) Model

2.1 Where Does Long Memory Come From?

Long memory in order flow arises from metaorder splitting. Institutional investors (funds, banks, market makers) want to build or unwind large positions, but cannot do so with a single massive order — that would move the market too strongly against them.

Instead, they split their large metaorder (e.g., "buy 100,000 shares") into many small child orders distributed over hours or days. Since all child orders have the same direction, the order flow appears predictable — but this is just the "echo" of the metaorder.

2.2 Power-Law Distribution of Metaorder Sizes

The LMF model assumes metaorder sizes Q follow a power-law distribution:

  • ψₘ(Q) ∝ Q^(−α−1), with 1 < α < 2

This links the memory parameters:

  • γ = α − 1, with 1 < α < 2

Empirically verified on the Tokyo Stock Exchange (TSE) and other markets.

3. The Square-Root Law of Price Impact

3.1 What Is Price Impact?

The price impact I(Q) of a metaorder of size Q is the total price change from the start to the end of the metaorder execution.

Linear impact (classical micro theory): I(Q) ∝ Q — the larger the order, the proportionally larger the price move. Intuitive, but empirically wrong for large orders.

Square-Root Law (SRL): Empirically:

  • I(Q) ∝ Q^δ, with δ ≈ 1/2

This means: price impact grows sublinearly — an order 4× larger only moves price ~2×. The nonlinearity is strictly universal: δ = 1/2 holds for all liquid stocks on the TSE and presumably in all developed markets.

3.2 Why Does This Matter?

The square-root law implies that large institutional investors move the market less than linear theory would predict. This is the mechanism that preserves Brownian price dynamics — even when order flow is predictable.

4. The Extended Nonlinear LMF Model (Sato & Kanazawa 2025)

4.1 Core Idea

The authors extend the LMF model with nonlinear price impact: each child order from trader i generates a price contribution of the form:

  • Δm^(i)(t) ∝ ε^(i) · (Q^(i))^δ

where δ ∈ (0, 1] is the impact exponent. For δ = 1/2 this is exactly the square-root law.

4.2 Mapping to Lévy-Walk Theory

The key mathematical step: the price contribution of a single trader can be mapped exactly to a Lévy walk with nonlinear walking speed.

A Lévy walk is a generalization of the random walk where "step lengths" follow a power-law distribution (rather than Gaussian). The model is exactly solvable — there is a closed-form analytical solution for the price dynamics.

4.3 Main Result: Brownian Diffusion Is Universal

The exact solution for the mean squared displacement (MSD) of price movement is:

E[Δm²(t)] ∝  t^(1+2δ−α)   if 2δ > α
              t              if 2δ ≤ α

Critical finding: Under the LMF assumption (1 < α < 2) and the square-root law (δ = 1/2):

  • 2δ = 1, and α > 1, so 2δ < α whenever α > 1
  • Therefore: E[Δm²(t)] ∝ t — normal Brownian diffusion, always

The square-root law sits exactly on the boundary between superdiffusion and normal diffusion. This is not a coincidence — it is the mechanism that enforces the EMH.

5. Phase Diagram: Diffusion vs. Superdiffusion

δ
1.0 |████████████████| Superdiffusion
    |████████████████|
0.5 |────────────────|← SRL boundary (δ = 1/2)
    |                | Normal Diffusion (Brownian motion)
0.0 |________________|
    1.0   1.5   2.0   α
  • Above the boundary (2δ > α): Superdiffusion — prices are predictable, arbitrage possible
  • On/below the boundary (2δ ≤ α): Normal diffusion — prices are Brownian motion, EMH holds
  • The square-root law sits exactly on the boundary → minimal nonlinearity of impact that still guarantees EMH consistency

6. Further Empirical Laws the Model Explains

6.1 Inverse Cubic Law of Price Changes

Empirically, large price changes follow:

  • P(Δm) ∝ (Δm)^(−β−1), with β ≈ 3

The model derives this analytically: β = α/δ = (3/2)/(1/2) = 3. No fine-tuning required.

6.2 Volatility Clustering

The model reproduces volatility clustering (periods of high/low volatility), measured by the covariance function of squared volatility:

  • Cᵥ(τ) ∝ τ^(−ζ), with ζ ≈ α − 1

This also emerges without artificial assumptions — it follows directly from metaorder dynamics.

7. Implications for Traders and Market Participants

7.1 Market Resilience Against Large Orders

The square-root law means: institutional investors can execute their metaorders without permanently destabilizing the market. The concavity of impact dampens large price moves — a natural stabilizing mechanism.

7.2 EMH on Long Timescales

Although order flow is predictable in the short term (visible through metaorder splitting), price movements are unpredictable on long timescales. Simple arbitrage strategies based purely on order flow fail — the SRL mathematically prevents it.

7.3 Practical Relevance for Execution

  • Slippage estimation: For large orders, use I(Q) ∝ √Q, not linear models. Underestimating slippage on large positions is a common and costly error.
  • Optimal execution: The concavity of the SRL incentivizes spreading orders over time (TWAP/VWAP) — which is exactly what institutional desks do.
  • Order flow as signal: Short-term order flow autocorrelation is real but does NOT translate to price predictability on timescales relevant to most traders.

7.4 Limits of the Model

  • No endogenous market dynamics (no self-reinforcement, no order book)
  • No time-varying parameters (no intraday seasonality, no external shocks)
  • No heterogeneous agents with different strategies
  • Single-asset focus, no cross-asset effects

These limitations make the model analytically elegant but too abstract for operational trading strategies alone. It explains the "why" of price dynamics, not the "when" of a specific trade.

8. Glossary

Term Meaning
Metaorder Large institutional order split into many small child orders
Child order Individual small sub-order of a metaorder
Order flow Time series of all buy/sell signals in the market
Long memory ACF decays slowly (power law), not exponentially
Square-Root Law (SRL) I(Q) ∝ √Q — nonlinear, sublinear price impact
Brownian motion Random diffusion: E[Δm²] ∝ t — unpredictable
Superdiffusion E[Δm²] ∝ t^η with η > 1 — predictable, arbitrage possible
Lévy walk Generalized random walk with power-law step lengths
EMH Efficient Market Hypothesis — prices reflect all public information
LMF model Lillo-Mike-Farmer model of market microstructure

9. Core Insight in One Sentence

The square-root law of price impact is not accidental — it is exactly the minimal nonlinearity that prevents predictable institutional order flow from producing predictable prices, thereby mathematically enforcing the Efficient Market Hypothesis on long timescales.