Skip to content
All library documents

Handling Sparse Trades in High-Frequency Return Features

Article Quant Q&A · Author: IGottaLearnMath
Read the original

Summary

The document raises a practical feature-construction issue when replicating research on high-frequency stock return predictability. It asks how to calculate transaction returns or price direction over very short intervals when trades may be much less frequent than the interval grid, and whether missing observations should remain missing, be set to zero, or be replaced with midpoint returns.

No solution or empirical comparison is provided; the text is a request for guidance prompted by a cited paper. It highlights that treating intervals without trades as ordinary observations could distort a training sample. Any implementation would need to define how to align trades and quotes, handle inactive intervals, and distinguish transaction-price changes from midpoint changes. The document does not establish which choice is best or report results, so it serves as a research question rather than a validated method.

Key ideas

  • Very short return intervals may contain no transaction when trading is sparse.
  • Filling inactive intervals with zero or missing values can affect model training.
  • Midpoint returns are raised as a possible alternative to transaction returns.
  • The document asks a replication question and provides no tested recommendation.

Tags

High-frequency tradingEquitiesMarket microstructureMachine learning
Full text
# Question on high frequency stock return predictability paper


# Question on high frequency stock return predictability paper












I'm trying to replicate some of the findings made in the paper below. I'm stuck and have a question on Page 9, where the transaction return or price direction features are being built.

Given the time intervals being discussed (.1, .2), (.2, .4), (.4, .8), in seconds, how do you deal with the frequent occurence of transactions not happening in that timeframe (if say the average trade time is 2secs). Leaving na or setting 0 to would surely skew the training data as their will be quite a few of those entries.

Should one just use the midpoint return as opposed to transaction return?

HOW AND WHEN ARE HIGH-FREQUENCY STOCK RETURNS PREDICTABLE? by Yacine Aït-Sahalia, Jianqing Fan, Lirong Xue, Yifeng Zhou, NBER Working Paper 30366

https://www.nber.org/system/files/working_papers/w30366/w30366.pdf

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

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