SwealthEdge

Explainers / DRIFT

Earnings drift

Price behaviour in the sessions after an earnings surprise, ranked by our own surprise measure. The effect this family is built on is the best-documented and most-decayed anomaly in the literature, and the page says so.

Engine family slug pead · methodology pead/0.3.1

What has to be true

  • An issuer reports results, and our surprise measure places the report in the top or bottom band for its sector over the trailing comparison period.
  • The session after the announcement closes outside the announcement-day range, in the direction the surprise implies.
  • A composite score is computed from the surprise percentile and the reaction, and the signal is published only when that score clears the family's threshold for that methodology version.
  • Both directions are detected and recorded. Downward cases are published as detections, in the same neutral language, and are not framed as anything to act on.

Thresholds

The settings this family applies. They are published because they are choices: a different window or a different count produces a different set of signals from the same data.

Surprise percentile
top or bottom band within sector

A percentile within the sector, not an absolute beat in pence or cents.

Composite score
must clear the threshold published on the row

The threshold travels with each signal in its evidence, so a change to it is visible on the rows it applied to rather than only in a changelog.

Window under test
20 trading days

The window the research measures over. It is not a holding recommendation.

Surprise input
time-series derived

Derived from the issuer's own history, NOT from point-in-time analyst consensus. See the limitations.

The published work it draws on

Named, not reproduced and not linked. Each entry says what that work reported, in its own terms. None of it was conducted by us, and none of it is a statement about the current market.

  • Ball, R. and Brown, P. (1968), ‘An Empirical Evaluation of Accounting Income Numbers’, Journal of Accounting Research 6(2).

    First documented that prices continued to move in the direction of an earnings surprise after the announcement rather than adjusting at once.

  • Bernard, V. and Thomas, J. (1989), ‘Post-Earnings-Announcement Drift: Delayed Price Response or Risk Premium?’, Journal of Accounting Research 27.

    Established the drift as a distinct effect and argued it was a delayed response rather than compensation for risk.

  • Chordia, T., Subrahmanyam, A. and Tong, Q. (2014), ‘Have capital market anomalies attenuated in the recent era of high liquidity and trading activity?’, Journal of Accounting and Economics 58(1).

    Reported that a number of documented anomalies, this one among them, weakened materially in more liquid markets and in later sample periods.

Where this does not work

Every detection method has conditions it was measured under and conditions it fails in. These are this one’s, written down rather than left out.

  1. This effect has attenuated. That is not our reading of it — it is the finding of the third paper above, and it is the single most important thing on this page. A measurement made on a 1970s or 1980s sample is not a description of today's market.
  2. Our surprise measure is derived from the issuer's own reporting history, not from point-in-time analyst consensus, because a clean point-in-time consensus dataset is a licensed product we do not display. A time-series surprise and a consensus surprise disagree often, and the published research was mostly built on the latter.
  3. The window under test is twenty trading days. That is a measurement choice inherited from the literature, and a different window produces a different ranking of the same events.
  4. Announcements happen outside trading hours. The prices a study measures from are frequently not prices a reader could have transacted at, and the gap is largest exactly where the measured effect is largest.
  5. Nothing here models commission, spread, borrowing cost for the downward cases, or the market impact of the position size a reader might choose.
  6. This family is at methodology version 0.3.1. The number is not modesty: this detector is younger than the insider-cluster one and its thresholds have moved more.

What this page cannot tell you

  • Whether an effect documented on historical samples is present in the current one. We have not established that.
  • What any individual should hold, buy, sell or avoid.
  • Anything about the quality of the business that reported.