Metaorder modelling and identification from public data

Battle Count: 52

The paper is highly relevant to quantitative trading in several ways: (1) It provides a method to reconstruct metaorders from public data, which is essential for calibrating market impact models used in execution algorithms. (2) The square root law and concave impact profiles are directly used in optimal execution strategies. (3) Understanding order-splitting behavior informs TWAP/VWAP algorithm design. (4) The LMF relation connects microscopic trading behavior to macroscopic price impact, relevant for predicting short-term price movements from order flow. (5) The finding that impact stylised facts alone don't guarantee LMF consistency is important for model validation. However, the paper is more foundational/theoretical than directly actionable for trading strategies.

The methodology involves multiple interconnected components: (1) Data preprocessing of high-frequency Level 1 data with specific cleaning rules. (2) Two algorithms for synthetic metaorder generation (mapping function and metaorder compilation). (3) Grid search over 50 parameter configurations per stock-year across 239 stocks × 3 years. (4) Multiple estimation procedures (NLLS, PSD, power-law fitting, runs test, FFT-based autocorrelation). (5) Variance-aware loss functions and aggregated error metrics. (6) Non-linear curve fitting for execution profiles and decay. The computational cost is significant due to the grid search, and the statistical methodology requires careful implementation of power-law estimation and autocorrelation analysis. The code is available but requires BMLL data access.

Code and notebooks are available in a public GitHub repository (https://github.com/EzraGoliath/Metaorder-modelling-and-identification-Msc-thesis-). However, the underlying high-frequency market data requires access to the BMLL Data Lab platform and is not redistributed. The repository is described as a 'moving code base' without an archived release, which limits long-term reproducibility. The methodology is well-documented with explicit algorithms and parameter grids.