Statistical Metaorders Analysis

• Statistical metaorders are sequences of coordinated same-sign market orders that reveal long-range autocorrelations and mechanical impact patterns.
• Reconstruction algorithms segment public trade data into synthetic metaorders, enabling precise analysis of power-law distributions and square-root impact laws.
• Empirical studies across asset classes confirm that market impact stems primarily from liquidity dynamics rather than informational advantages.

A statistical metaorder is a sequence of same-sign (all buy or all sell) market orders that are executed in a coordinated manner—typically by a single trader or strategy, but possibly reconstructed synthetically from transaction data—such that the collective action exhibits a substantial, measurable effect on market prices. The study of statistical metaorders is motivated by empirical findings that market order flow displays long-range autocorrelations, and that large trading intentions are usually implemented as a sequence of smaller transactions (order-splitting), giving rise to characteristic market impact laws and persistent flow correlations. The reconstruction and quantitative modeling of statistical metaorders enable rigorous analysis of their distributional properties, market impact, and the underlying mechanism generating long memory in order flow, even when proprietary order-level identifiers are not available.

1. Theoretical Framework: Order-Splitting and Long-Memory

The foundational hypothesis for statistical metaorders derives from the Lillo–Mike–Farmer (LMF) order-splitting theory. The theory posits that observed long-range autocorrelations in the sign of market orders,

$$C(\tau) = \langle \epsilon_\ell \, \epsilon_{\ell+\tau} \rangle \sim \tau^{-\gamma}$$

for large $\tau$, are a direct consequence of strategic traders splitting large metaorders into sequences of same-sign child orders. According to LMF, metaorder lengths $L$ follow a power-law distribution $P(L) \sim L^{-(\alpha+1)}$ with $\alpha > 1$, and the decay exponent $\gamma = \alpha - 1$ exactly links the distribution of metaorder lengths to the autocorrelation of order signs:

$$P(L) \sim L^{-(\alpha+1)}, \quad C(\tau) \propto \tau^{-\gamma}, \quad \gamma = \alpha - 1$$

Independence among traders’ submission times allows separation of their contributions, yielding a superposition principle for the aggregate flow (Goliath et al., 23 Feb 2026).

2. Algorithmic Reconstruction from Public Data

Due to lack of trader identifiers in standard public data (e.g., TAQ), Maitrier et al. (Maitrier et al., 23 Mar 2025, Goliath et al., 23 Feb 2026) and subsequent works developed robust algorithms to reconstruct “synthetic” metaorders by mapping trades to fictitious agents and segmenting each agent’s time series into metaorders (consecutive same-sign trades).

The standard workflow proceeds as follows:

1. Synthetic Trader Assignment: Assign each trade in chronological order to one of $N$ synthetic traders using a multinomial mapping driven by a participation distribution $F(f)$.
2. Metaorder Segmentation: For each synthetic trader, group consecutive trades of the same sign into metaorders. Each block of $n_j\geq2$ same-sign trades forms a single synthetic metaorder.
3. Feature Extraction: For each metaorder $\tau$0, extract the number of child orders $\tau$1, total volume $\tau$2, start/end prices, impact $\tau$3, duration, daily volume $\tau$4, and intraday volatility $\tau$5.

The precise choice of $\tau$6 and $\tau$7 affects only the coarseness of reconstructed metaorders but does not materially alter key statistical exponents or stylized facts when $\tau$8 is set in a plausible range (e.g., $\tau$9 or $L$0) (Goliath et al., 23 Feb 2026).

3. Stylized Facts: Empirical Laws of Metaorders

Systematic empirical analysis using both proprietary metaorder data and public-data-reconstructed synthetic metaorders robustly reproduces several statistical laws:

• Metaorder Length Distribution: Power-law tails with exponents $L$1 are universal, confirmed for equities, futures, options, and cryptocurrencies (Goliath et al., 23 Feb 2026, Said et al., 2018, Said et al., 2019, Donier et al., 2014).
• Square-Root Impact Law: Peak price impact $L$2 scales as

$L$3

over four decades in $L$4, insensitive to order duration or execution style (Maitrier et al., 23 Mar 2025, Donier et al., 2014, Said et al., 2019).

• Temporal Independence: Impact is empirically independent of the metaorder’s duration $L$5, provided $L$6 exceeds a minimal threshold, confirming the role of cumulative executed volume as the main variable.
• Concave Execution Profile: The impact during execution grows as a concave function of the fraction executed, $L$7 with $L$8 for liquid instruments (Goliath et al., 23 Feb 2026, Maitrier et al., 23 Mar 2025).
• Convex Post-Trade Decay: Impact decays after completion according to propagator models with kernel exponent $L$9, leading to

$P(L) \sim L^{-(\alpha+1)}$0

for $P(L) \sim L^{-(\alpha+1)}$1 (Goliath et al., 23 Feb 2026, Bucci et al., 2019).

• Permanent Impact Plateau: Empirically, the permanent price effect stabilizes at approximately $P(L) \sim L^{-(\alpha+1)}$2 of the peak immediate impact, matching theoretical predictions under fair-pricing conditions and power-law metaorder size distributions (Farmer et al., 2011, Said et al., 2018, Said et al., 2019).

These relations persist across assets, venues, and market regimes, including limit and market order executions, options, and digital assets (Goliath et al., 23 Feb 2026, Said et al., 2019, Donier et al., 2014).

4. Statistical Estimation and Exponent Linkages

The recovery of LMF’s key scaling $P(L) \sim L^{-(\alpha+1)}$3 can be operationalized by maximum likelihood estimation of the metaorder-length tail exponent $P(L) \sim L^{-(\alpha+1)}$4 (using the Clauset–Shalizi–Newman MLE for power laws) and log–log regression of the sign autocorrelation function decay exponent $P(L) \sim L^{-(\alpha+1)}$5 (Goliath et al., 23 Feb 2026). Empirical measurements on large stock universes (e.g., top 100 JSE stocks, Russell 3000) unambiguously confirm that $P(L) \sim L^{-(\alpha+1)}$6, with confidence intervals ruling out alternative explanations for the majority of instruments.

Furthermore, alternative theoretical frameworks—such as anticipative price formation under Hawkes processes (Jaisson, 2014), and propagator models with mechanical liquidity response—derive the square-root law and long-memory as necessary consequences of basic flow and price-formation postulates, not requiring real-time detection of metaorders (Maitrier et al., 9 Jun 2025, Maitrier et al., 5 Sep 2025).

5. Mechanical versus Informational Impact

Empirical analysis of both synthetic and proprietary metaorders decisively shows that the observed short-term impact laws, including the square-root law and impact decay, can be fully recovered by random (uninformed) synthetic metaorders constructed solely from the anonymized flow. Randomizing trade assignments removes all informational edge, yet impact profiles and exponents remain unchanged (Maitrier et al., 23 Mar 2025, Goliath et al., 23 Feb 2026).

This implies the average realized impact is predominantly mechanical, arising from the endogenous interaction between aggregate market order flow and latent order book liquidity replenishment. Informational effects, such as market response to news, appear as secondary corrections. These findings are reinforced by analyses showing that for statistically isolated (uninformative) metaorders, the mechanical impact decays to zero, whereas for correlated/informed metaorders, the permanent component remains (Donier et al., 2014, Bucci et al., 2019). As a result, universality of the square-root impact law and stability of metaorder exponents are attributed to flow–liquidity mechanics, not agent information (Maitrier et al., 23 Mar 2025, Goliath et al., 23 Feb 2026).

6. Broader Implications and Methodological Significance

The ability to reconstruct statistical metaorders from public data eliminates a major obstacle to market microstructure research, namely the historic dependence on proprietary datasets with trader or broker identifiers. The developed mapping and segmentation algorithms now enable fully reproducible, cross-market studies of metaorder impact, long memory, and liquidity dynamics (Goliath et al., 23 Feb 2026, Maitrier et al., 23 Mar 2025).

This methodological advance supports a wide array of research, including agent-based models and reaction–diffusion models of latent liquidity, and aligns robust empirical results with reaction–diffusion and order-driven theories of market impact and volatility (Maitrier et al., 9 Jun 2025, Maitrier et al., 5 Sep 2025). It also allows rigorous validation of theoretical predictions, such as the fair-pricing equilibrium and the link between metaorder size distributions and observable flow correlations, across global asset classes (Farmer et al., 2011, Said et al., 2019, Goliath et al., 23 Feb 2026).