Paper: arXiv 2405.10654
Authors: Salma Elomari-Kessab, Guillaume Maitrier, Julius Bonart, Jean-Philippe Bouchaud
Abstract
Understanding the micro-dynamics of asset prices in modern electronic order books is crucial for investors and regulators. In this paper, we use an order by order Eurostoxx database spanning over 3 years to analyze the joint dynamics of prices and order flow. In order to alleviate various problems caused by high-frequency noise, we propose a double coarse-graining procedure that allows us to extract meaningful information at the minute time scale. We use Principal Component Analysis to construct “microstructure modes” that describe the most common flow/return patterns and allow one to separate them into bid-ask symmetric and bid-ask anti-symmetric. We define and calibrate a Vector Auto-Regressive (VAR) model that encodes the dynamical evolution of these modes. The parameters of the VAR model are found to be extremely stable in time, and lead to relatively high \(R^2\) prediction scores, especially for symmetric liquidity modes. The VAR model becomes marginally unstable as more lags are included, reflecting the long-memory nature of flows and giving some further credence to the possibility of “endogenous liquidity crises”. Although very satisfactory on several counts, we show that our VAR framework does not account for the well known square-root law of price impact.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced statistical techniques like PCA and VAR modeling with multi-lag stability analysis, while being grounded in a specific, large-scale 3-year dataset with detailed pre-processing and reported prediction metrics.
Research Flowchart
flowchart TD
A["Research Goal:<br>Disentangle joint dynamics of<br>prices & order flow"] --> B["Data Input:<br>3y Eurostoxx order-by-order data"]
B --> C["Methodology:<br>Double Coarse-Graining<br>to minute scale"]
C --> D["Computational Process:<br>Principal Component Analysis<br>extracts 'Microstructure Modes'"]
D --> E["Modeling:<br>Vector Auto-Regressive Model<br>calibrated on modes"]
E --> F{"Key Findings"}
F --> G["VAR parameters stable<br>High R² for symmetric liquidity"]
F --> H["Model marginally unstable<br>(Long memory/Endogenous risk)"]
F --> I["Model fails<br>Square-root impact law"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a double coarse‑graining procedure to extract minute‑scale information and defines a calibrated Vector Auto‑Regressive (VAR) model for the microstructure modes.
Source sentence
we propose a double coarse-graining procedure that allows us to extract meaningful information at the minute time scale. We
Assumes 6
- EuroStoxx is a large‑tick, liquid asset with a spread almost always equal to one tick. market
Source sentence
It is noteworthy that EuroStoxx is a liquid large-tick asset, with a spread almost always equal to one tick.
- After binning, null flows disappear, making the VAR model applicable. data
Source sentence
... an effect that completely disappears when the data is binned.
- The linear VAR framework is suitable for modeling the dynamics of the transformed variables. model
Source sentence
We adopt a Vector Autoregression (VAR) model to capture the dynamic relationships among the variables associated with each price change.
- For short time scales, market orders issued by slow traders have little short‑term predictability, allowing the second term in the impact definition to be neglected. execution
Source sentence
for short enough time scales, market orders issued by slow traders have little short term predictability such that the second term in Eq. (11) is negligible.
Show 2 more
- Box‑Cox transformations render the binned variables approximately Gaussian. statistical
Source sentence
These parameters are chosen to maximise the likelihood of the Gaussian distribution of the transformed variables, which yields λv = 0.20 and λt = 0.14.
- The impact curves from prior literature can be used to compute instantaneous returns within the VAR framework. model
Source sentence
due to the use of the impact curves from [22] for computing the instantaneous return, the return produced by our model is diluted in scale.
Data
- EuroStoxx futures order‑by‑order database order-book unknown
Source sentence
The dataset used in this study consists of 545 days of the futures contract on EuroStoxx from September 2016 to August 2019.
Costs
Not applicable
Finds 5
- The VAR model parameters are extremely stable over time and achieve relatively high R² prediction scores, especially for symmetric liquidity modes (≈0.28–0.32). predictive-accuracy
Source sentence
R2 scores are much higher (∼0.28 −0.32) for symmetric modes, which carry no information on returns, than for anti-symmetric modes (∼0.01 −0.03).
- In‑sample and out‑of‑sample R² scores are close, indicating statistically significant predictive value of the VAR model. statistical
Source sentence
the in-sample R2 scores and out-of-sample scores are close, highlighting the fact that the predictive value of the VAR model is statistically significant.
- Adding more lags (up to 10) increases out‑of‑sample R² for all modes by about 25% but makes the system marginally less stable. predictive-accuracy
Source sentence
adding more lags reduces auto‑correlation of residuals and increases the out of sample R2 score of all the modes, by ∼25% both for the symmetric and anti‑symmetric ones when p increases from 1 to 10.
- The VAR framework does not reproduce the well‑known square‑root law of price impact. negative
Source sentence
our VAR framework does not account for the well known square‑root law of price impact.
- Flow variables improve return prediction: VAR yields a 1.6% R² for returns, higher than the 0.49% obtained using past returns alone. predictive-accuracy
Source sentence
... significantly higher than the score of 0.49% obtained when predicting returns using the past return as the only feature.
Checked by others
Contradicted by
- The two square root laws of market impact and the role of sophisticated market participants
Compare the findings
This paper: The VAR framework does not reproduce the well‑known square‑root law of price impact.
That paper: During execution, market impact follows a square‑root law in the cumulated traded volume for a given participation rate.
5 papers in the archive : Why is the estimation of metaorder impact with public market …
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (32)
- The Impact of Market Orders (2018)
- Universal scaling and nonlinearity of aggregate price impact in financial markets (2018)
- Generative AI for End-to-End Limit Order Book Modelling: A Token-Level Autoregressive Generative Model of Message Flow Using a Deep State Space Network — → 2309.00638
- New introduction to multiple time series analysis (2005)
- Impact is not just volatility (2019)
- Crossover from linear to square-root market impact (2019)
- Learning to simulate realistic limit order book markets from data as a World Agent (2022)
- A generative model of a limit order book using recurrent neural networks (2023)
and 24 more
- Cross impact of order flow imbalances: Contemporaneous and predictive (2021) — link
- Conditional generators for limit order book environments: Explainability, challenges, and robustness — → 2306.12806
- Towards realistic market simulations: a generative adversarial networks approach (2021)
- Stock price jumps: news and volume play a minor role (2008)
- Exogenous and endogenous price jumps belong to different dynamical classes (2022)
- Endogenous versus exogenous origins of crises (2006)
- Critical reflexivity in financial markets: a Hawkes process analysis (2013)
- Branching-ratio approximation for the self-exciting Hawkes process (2014)
- The endogenous dynamics of markets: Price impact, feedback loops and instabilities (2011)
- How markets slowly digest changes in supply and demand (2009)
- The predictive power of zero intelligence in financial markets (2005)
- Riding Wavelets: A Method to Discover New Classes of Price Jumps — → 2404.16467
- A fully consistent, minimal model for non-linear market impact (2015)
- Slow decay of impact in equity markets: insights from the ANcerno database (2018)
- Why is equity order flow so persistent? (2015)
- Statistical analysis of zero-inflated nonnegative continuous data (2019)
- Hawkes processes in finance (2015)
- Price impact. Encyclopedia of Quantitative Finance. 2010
- Handbook of Price Impact Modeling (2023)
- Endogenous liquidity crises (2020)
- Master curve for price-impact function (2003)
- Anomalous price impact and the critical nature of liquidity in financial markets (2011)
- Modeling zero-modified count and semicontinuous data in health services research Part 1: background and overview (2016) — link
- Analyzing Skewed Continuous Outcomes With Many Zeros: A Tutorial for Social Work and Youth Prevention Science Researchers (2018) — link
3 of 32 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.