Estimating Aggressive Trading from Price Changes and Return Distributions

Summary

This research note compares ways to estimate buyer- and seller-initiated trading. Tick-based classification uses trades relative to the prior best bid and ask, while a bar-based flow measure assigns volume according to the direction of the close-to-close move. A batch classification method also incorporates the size of the price change, mapping larger moves to higher estimated aggressive-trade shares. The note then estimates buying volume using standardized returns or return ranks paired with t, normal, or uniform distributions, and forms a factor from estimated buying volume as a share of total volume.

The reported tests find stock-selection information across the broad market and the CSI 800; return-rank versions paired with normal or uniform distributions are described as more stable. Nonlinear transformations of the factor’s upper-ranked region reportedly improve the stated information coefficients. The note also says the factor overlaps with reversal and volatility effects: after style neutralization, excess-return ability disappears, though long-short returns remain. The extract gives no full test setup or detailed performance period, so the metrics should be read in that context.

Key ideas

• Trade direction can be approximated from price changes relative to quotes or prior closes.
• Batch classification estimates aggressive buying share using both direction and the magnitude of price moves.
• The study tests distribution-based mappings from standardized returns or return ranks to estimated buying volume.
• Return-rank versions using normal or uniform mappings are reported as more stable in the tested universes.
• The factor is related to reversal and volatility, and style neutralization removes its excess-return ability.

Tags

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