Post-earnings-announcement drift
Prices have historically kept drifting in the direction of an earnings surprise for weeks after the report — one of the most durable anomalies in the academic literature. Finisdom measures whether that held for the names you follow, using the original seasonal-random-walk formulation on filings pulled straight from the SEC, and shows the base rates with their caveats attached rather than a signal.
What it does
- Every reported quarter since 2008, pulled from SEC XBRL filings — not a four-quarter API window
- Surprise measured against the same quarter a year earlier, in the company’s own standard deviations
- Forward drift at 5, 21, and 63 sessions, net of the unconditional drift
- Sample size shown on every cell — the tails are always the thinnest
- The original Foster/Bernard-Thomas formulation: no analyst consensus required
- States plainly that the filing date is a proxy for the announcement date
How it works
Bucket the surprise
Each quarter is scored against the same quarter a year earlier and standardised by that company’s own history of year-over-year changes, then placed in one of five fixed buckets from big miss to big beat.
Measure what followed
Forward returns at 5, 21, and 63 sessions from the filing, pooled across the whole universe — with the unconditional drift subtracted so you see what the surprise added.
Read it with the sample
Every number carries its n. A bucket with a handful of events is a curiosity; one with a hundred is evidence.
Frequently asked
What is post-earnings-announcement drift?
The tendency for a stock to keep moving in the direction of an earnings surprise for weeks after the announcement, rather than repricing instantly. It was first documented in the late 1960s and has been studied ever since as one of the more persistent challenges to strict market efficiency.
How is the surprise measured without analyst estimates?
By comparing each quarter to the same quarter a year earlier and dividing by that company’s own history of year-over-year changes — a seasonal random walk. This is the original academic formulation (Foster, Olsen & Shevlin 1984; Bernard & Thomas 1989), and it needs no analyst consensus. It asks whether the business changed, not whether the street was wrong.
Why is a 75% EPS jump sometimes shown as "in line"?
Because the surprise is measured in that company’s own standard deviations, not in percent. For a business whose earnings swing wildly, a large jump is normal and scores near zero; for a steady one, a small beat can be a big surprise. That is the standardisation working as intended.
Is this a trading signal?
No. It reports what happened after past reports that looked similar, with the sample size and the caveats attached — that earnings cluster in a few weeks each quarter so events are not independent, and that the SEC filing date is a proxy for the announcement date. It never predicts what any specific name will do next.
Why do the numbers look modest?
The universe here is large, liquid, heavily-covered companies. The classic drift effect was always strongest in smaller, less-covered names, so this is a deliberately conservative read rather than a flattering one.
More Finisdom tools
Backtest any multi-asset portfolio against decades of real history — Sharpe, Sortino, Calmar, max drawdown, rolling returns, and crash stress tests, benchmarked and split by bull and bear regimes.
Trace the efficient frontier for any set of assets and compare max-Sharpe, min-volatility, equal-weight, inverse-vol, and risk-parity allocations — with honest walk-forward validation, not a curve fit.
Decompose any portfolio’s return against its benchmark — allocation, selection, and interaction effects — with factor-risk exposures, tracking error, information ratio, and marginal contribution to risk (MCTR).
A single 0–100 read on whether the market backdrop favours deploying capital or holding back — built from trend, credit, volatility, and breadth signals, with an S&P 500 deployment-zone overlay.
Pair a disciplined quant signal with a Claude fundamental-quality review, blended into one verdict — momentum, value, and quality scored by the engine, business durability read by the model.
Test a whole library of rule-based timing strategies on any instrument through a robustness funnel — an out-of-sample split, six pass/fail filters, and a permutation “skill vs luck” test — to see how few actually survive.
Real-money implied probabilities from Kalshi prediction markets for the Fed, inflation, recession, and GDP — plus a Claude-written weekly brief that reads the whole macro picture in plain English.
Track a dozen economically-linked pairs — gold vs real yields, credit vs stocks, the 60/40 hedge — and see which have pulled apart from their own norm, each with the historical base rates of what followed.
Ten concentrated, actively-managed funds — their largest positions, what they opened, exited, added and trimmed last quarter, and where any of it overlaps your own book. Index giants deliberately excluded.
Design a fund as a rule rather than a basket, replay it point-in-time across decades, then freeze it and publish it to a leaderboard that ranks funds on how much of their backtest they actually kept.
Backtests and optimizations are hypothetical illustrations built from historical data — not predictions, and not investment advice. Past performance does not guarantee future results.