Combining Order-Book Intent and Trade Data for Stock Selection

Summary

This research summary describes combining order-book data, which may reflect traders’ unexecuted intentions, with transaction-level data showing completed trades. The authors use both intuitive signal construction and machine learning to turn different types of high-frequency data into lower-frequency stock-selection factors. They report that buy-intent share and intraday buy-intent intensity measures show monthly predictive ability, and that some machine-mined factors can be interpreted through patterns in intraday buying and trading activity.

The reported evidence includes information coefficients, information ratios, win rates, and long-short returns; the summary also says several factors improved a CSI 500 enhanced portfolio. These are findings as presented in the supplied abstract, not independently verified results. The document does not provide the full paper, detailed factor definitions, sample and testing design, transaction-cost treatment, or out-of-sample evidence, so practical robustness cannot be judged from this excerpt alone.

Key ideas

• Order-book submissions and completed trades capture different aspects of market participants’ buying activity.
• Combining these data sources can produce lower-frequency stock-selection factors from high-frequency observations.
• The summary reports predictive results for buy-intent share and intraday buy-intent intensity measures.
• Some machine-mined factors are linked to interpretable patterns in buying and trading activity.
• The abstract reports improvements to an enhanced equity portfolio, but the supplied excerpt lacks enough testing detail to assess robustness.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.