Post-Earnings Drift by Sector, Walked 220 Times Under Two Anchors: Dead at the Formation Date, Alive at the Announcement

Every result in this paper comes from one rule set, run twice. The two runs differ in the day a position is opened, and in the two things that day forces: how a surprise qualifies, and how many names the book holds at once. Everything else, the universe, the surprise measure, the sector map, the costs, the floors, is identical.

What a surprise is. When a company reports, take actual earnings per share minus the consensus estimate. Divide that gap by how widely the company's own last eight surprises have scattered. The result is the standardized surprise, SUE. A two-cent beat counts as large for a company that usually lands within a cent of consensus, and as small for a company that misses by a dime every quarter. This step looks only at the company's own history; a company with fewer than four past surprises has no score and cannot be traded. A price-scaled surprise was rejected during the data pull: the vendor restates each name's EPS history by its own split factor, so scaling by price mixes bases across names, while SUE cancels the factor row by row because actual and consensus share it.

When the number is known. A report published after the close is known the next session. A report published before the open is known the same session. Every entry in this paper happens at the close of the session the number is known, so the overnight gap and the first day's reaction are always excluded.

Run one, the quarterly book. On the first trading day of January, April, July and October, take every company in the sector that reported inside the previous 90 days and rank them by SUE. Buy the five highest, short the five lowest, equal weight, long side and short side the same size. Hold for the fixed holding period, 5, 10, 21, 42 or 63 trading days, then close everything and sit in cash until the next quarter start. This is the design the academic literature and the factor products use. By the time the book forms, most of its names reported six to ten weeks earlier.

Run two, the announcement book. No fixed formation day. Each company is judged alone, on the session its number is known. Its SUE is compared with the pool of every SUE the sector printed in the last 365 days, roughly the last four reporting seasons, every one of them already on the record. Above the pool's 80th percentile, buy at that session's close. Below the 20th, short at that close. Anything in between does nothing, the sign has to agree with the side, and a surprise of exactly zero never enters. Because the cutoff comes from the past year, it is known on the day and does not wait for the rest of the season. Hold for exactly the holding period, then close. One open position per company; each new position takes a tenth of book equity, shared equally when more than ten are open on one side.

The consequence of the second rule is that the book is not five against five. In a season where many companies beat by more than last year's standard, the long side fills with more names than the short side. Technology is the sector where that happens most, and Table 6 counts it.

Why the columns of the tables are separate books. The five holding periods are five independent runs of the same rule. They enter the same names on the same days and differ in the day they exit; at the longer holds a repeat report that lands while a position is still open is skipped, so those columns carry slightly fewer entries. The 10-day column and the 63-day column of Technology hold mostly the same companies; one lets go after two weeks and the other after a quarter.

What a window is. Each book is scored over one calendar year at a time, twenty years in a row, 2006 through 2025, each year sealed before it is run. A year with fewer than fifteen different reporting companies in the sector is refused and printed as excluded rather than scored on a handful of names; the quarterly book applies the same fifteen-name floor at each formation day.

Costs and returns. Each traded name pays its liquidity tier one way at entry and at exit: 1.5 basis points above 500 million dollars of average daily volume, 3 above 100 million, 6 below. Shorts pay 50 basis points a year of borrow. The announcement long-short book's realized drag averaged 0.66 percent per one-year window. Returns are total returns, dividends credited from the payment record on their ex-dates, never from the vendor's adjusted close.

The record. All 220 studies, 11 sectors by 5 holding periods by 2 book shapes by 2 entry rules, were registered before any result was read; each year is a sealed hypothesis with a frozen run report, and every number in this paper is re-derived from those frozen reports, never from walk logs. The study code carries 14 unit tests including a hand-computed P&L check; rehearsals on unregistered cells caught three defects that were fixed before registration; the two runs took about eleven hours of engine time with no failures, and the family of 220 was declared and stamped into every artifact before this text was written.

The explorer above draws all 220 books; the tables below score them. The two long-short grids are Tables 1 and 2, and the whole result sits in their difference. Formed quarterly, the grid averages -0.5 percent a year net; eight of the eleven sector rows average below zero, Industrials sits at +0.02, and only the two short-record sectors average above. The worst cells sit a month and further out: Technology from 21 days and Financials at one to two months, each losing about 3 percent a year. Entered at the announcement, the same design averages +2.2, and 35 of the 55 cells print positive means. The average cell improves by 2.7 points on the switch; technology improves by 7.7 to 13.5 points depending on the rung.

The holding-period profile (Table 7, and the second figure) is the mechanism's signature. At the announcement the grid mean rises from +2.3 at five days to +3.1 at ten, holds +2.9 at a month, then decays to +2.2 at two months and +0.2 at the quarter. The formation profile sits between -0.2 and -0.9 at every rung: by the time a quarter-start basket forms, the window Bernard and Thomas measured has mostly closed, and what remains to harvest rounds to zero or less after costs.

What the ten-day rung survives in costs: its +3.1 is net of tiers that drag 0.7 points a year, so the gross is +3.8 and spreads would need to reach about 5 times the charged schedule, roughly 8 basis points a side for a top-tier name and 33 for the bottom tier, before the rung rounds to zero.

Technology is the loudest single story (Table 6). At 21 days: -3.0 percent a year formed quarterly, +10.5 entered at the announcement, 80 percent of twenty windows positive, and the ten-day and 42-day rungs agree at +9.8 and +8.2. Industrials at 21 days prints +7.8 with 85 percent of windows positive. Consumer Cyclical peaks at +7.7 at ten days. Energy is positive at all five announcement rungs. Financials and Healthcare hold gains at the one-to-two-week rungs and give them back by the quarter.

Utilities is the control group nature provided: negative at every rung under both anchors, -0.7 to -3.0 announcement-entered and -0.3 to -2.2 formed. Regulated cash flows leave little for a surprise to predict, and the sector prices what little there is inside the first week. Communication Services needs its label read with the map. The dated map counts each name by its label on the day, so before the September 2018 redraw the row is the telecom and media names that were never relabeled, a different book by composition. The fifteen-name floor admits that book from 2013 in the announcement run and only from 2019 in the quarterly run, which is why Table 1 carries 13 windows and Table 2 carries 7. The 7-window formation column is the best-looking in Table 2; the 13-window announcement row walks the same label back to noise and a -5.5 quarter rung. Both stand on few windows, and both are printed.

The long-only pair inverts on the same switch (Tables 3 and 4). Formed quarterly, picking the five largest surprises adds +0.2 points a year over holding every eligible name, and Industrials at the quarter rung is the cleanest cell in either wave: +5.1 points, 18 of 20 windows. Entered at the announcement, the same selection averages -1.5 points against a book that holds every announcement, because that book already captures the timing. Move the anchor and the Industrials edge disappears: it was timing, recorded under a selection heading.

Attribution. The announcement legs float with the news flow, so the book is net long when beats outnumber misses. Across 88,859 positions the mean tilt is +0.05 and the per-window correlation between tilt and return is 0.12: the grid-level result does not ride market exposure. Technology is the exception worth naming: 7,355 longs against 4,648 shorts (tilt +0.23, Table 6), because beats dominate the sector and a zero surprise never enters. Part of the technology row is net-long exposure in a rising market; the ten-day peak and the decay profile are not, and no other sector's tilt exceeds 0.12.

The literature said both of these things; it just said them fifty years apart. Ball and Brown found the drift in 1968 on announcement-relative time. Bernard and Thomas measured its decay inside sixty days in 1989. Chan, Jegadeesh and Lakonishok carried it into portfolio practice in 1996 on monthly formation, and the factor industry inherited that calendar. Martineau, reading modern large caps on formation-style designs, declared the drift dead in 2021. This grid holds all of it at once: the formation wave reproduces the disappearance, the announcement wave reproduces the original finding, and the rung profile shows why both are measurements of the same object. The anomaly did not vanish; the market's absorption of the number got faster than the quarterly calendar built to harvest it.

The sector map adds the part the pooled literature averages away. Three camps, declared as the design intended: drift that survives immediate entry (Technology, Industrials, Consumer Cyclical, Energy, Financials at the short rungs), a sector that reverses under every anchor (Utilities), and rows where nothing survives costs in either frame (Consumer Defensive, Basic Materials, Real Estate, and the young Communication Services record). A pooled index book blends all three and reports the average, which is how a live effect and a dead one can share a decade of literature.

What a reader can check: every one of the 220 cells is a page on this platform with its sealed prospectus, its per-window reports, its exclusions printed, and the family of 220 in the footer. This page is the announcement-entered Technology cell at 21 days; its formation twin is registered under the same title without the anchor clause, and the grid tables above name every other cell.

One era: 2006 through 2025, one bull-heavy market regime, and the two short-record sectors carry 13 and 10 announcement windows against everyone else's 20, with the Communication Services record before the September 2018 redraw made of the never-relabeled telecom and media names, admitted from 2013 in the announcement run and 2019 in the quarterly; Real Estate enters from 2016 and 2017 the same way. Daily closes only: entry is the close of the first session the announcement is known, so the overnight gap and the first day's reaction are excluded by construction, which understates what faster execution captures and is the version a daily-data reader can reproduce. Costs are tiered spreads without market impact, fair for five-name large-cap books and untested at size; borrow is a flat 50 basis points and ignores scarcity. One surprise measure by design; other definitions were not run and are not claimed. The consensus history is one vendor's record. The announcement long-short book floats its legs, and Technology's +0.23 tilt means part of that row is market exposure, quantified in Table 6 and the attribution paragraph. Exclusions are concentrated where the design predicted, in the two young sectors before their redraw dates, and every one is printed on its study page.

Where the story stops: the grid is measured, and nothing here has been deployed. The one open check that could move the headline is a beta-neutral rerun of Technology, whose announcement legs tilt +0.23 long; no other sector's tilt exceeds 0.12, so the rest of the grid does not wait on it. If a hedged Technology row keeps most of its +10.5, the sector story stands alone; if it does not, the row is partly a market bet and this paper has already said which part.

1. Ball, R., & Brown, P. (1968). An empirical evaluation of accounting income numbers. Journal of Accounting Research, 6(2), 159-178. (First documentation of drift in the direction of the earnings news after the announcement.)
2. Bernard, V. L., & Thomas, J. K. (1989). Post-earnings-announcement drift: delayed price response or risk premium? Journal of Accounting Research, 27, 1-36. (The drift concentrated in the weeks after the announcement, fading within roughly sixty days.)
3. Bernard, V. L., & Thomas, J. K. (1990). Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. Journal of Accounting and Economics, 13(4), 305-340. (The predictable pattern of returns around subsequent announcements.)
4. Chan, L. K. C., Jegadeesh, N., & Lakonishok, J. (1996). Momentum strategies. Journal of Finance, 51(5), 1681-1713. (Earnings momentum on formation-style portfolios; the horizon over which it pays.)
5. Chordia, T., Goyal, A., Sadka, G., Sadka, R., & Shivakumar, L. (2009). Liquidity and the post-earnings-announcement drift. Financial Analysts Journal, 65(4), 18-32. (The drift concentrated in illiquid names once trading costs are charged.)
6. Martineau, C. (2021). Rest in peace post-earnings announcement drift. Critical Finance Review (forthcoming; SSRN 3111607). (Price discovery of earnings news accelerated; drift in large caps shrank toward zero in recent decades on standard designs.)