Data-Driven Measures of High-Frequency Trading

Battle Count: 80

Highly relevant to quantitative trading. The paper provides publicly available daily measures of liquidity-supplying and liquidity-demanding HFT for all U.S. stocks (2010-2023), which can be directly used in trading strategy development, execution algorithm design, liquidity forecasting, and market impact modeling. The findings on how HFT composition affects spreads, information acquisition, and informed trading are directly actionable for quant traders designing strategies around liquidity provision, adverse selection, and information asymmetry. The validation against exchange technology changes and latency arbitrage provides practical insights for trading infrastructure decisions.

Moderate complexity. The ML training requires proprietary Nasdaq HFT data and access to WRDS TAQ Intraday Indicators. However, the authors make the final stock-day measures publicly available, reducing implementation burden for end users. The 2SLS IV estimation is standard econometrics. Reproducing the full pipeline requires: (1) access to Nasdaq HFT data for training, (2) WRDS TAQ data for features, (3) tree ensemble implementation (available in scikit-learn), (4) IV estimation with proper clustering. The validation exercises require additional proprietary data (Euronext, LSEG Datascope).

The ML-generated HFT measures are made publicly available for researchers. WRDS TAQ data is used for input features to improve replicability. However, the training data (Nasdaq HFT dataset) is proprietary and not publicly available. The Euronext validation data (Eurofidai-Bedofih) is also proprietary. The methodology is well-documented with detailed variable definitions and estimation procedures.