Shorting bad earnings: post-earnings-drift backtest loses −1.48% per trade

After a bad earnings report, a stock often doesn't drop in one move — it keeps sliding for weeks. That's the post-earnings-announcement drift, and in theory a trader shouldn't need to dive into the chaos of the earnings day itself to profit from it: waiting for the first reaction and then shorting should be enough. This preregistered backtest ran exactly that rule on 8,861 short trades across 2,615 different small-cap U.S. stocks, from 2011 through 2026.

Small caps are supposed to be the last place where the drift survives — in large, heavily-followed stocks, the academic literature says it has largely vanished. Only the short side is tested here: sell after the market has already reacted to the bad numbers, then cover 20 trading days later. Entering into the earnings-day gap itself, and buying after good earnings, are both outside the scope of this study.

The result in four numbers

• −1.48% — the average result of a short trade after all costs, using the winsorized mean (a mean where the most extreme 1% of values on each end are capped before averaging, so a few outliers can't dominate the number). Across 8,861 trades, this rule cost money — it didn't make any.
• −2.08% to −0.87% — the range the result falls in with 95% confidence (a bootstrap confidence band: a range built by repeatedly resampling the data itself, showing how stable an average really is). It sits entirely below zero — this isn't a fluke.
• 3 of 15 years were profitable at all, and 0 of 4 time periods. There is no stretch of history in which the rule worked — it wasn't good once and dead now; it was underwater the entire way through.
• −1.15 percentage points — how much worse the short after bad earnings performed than the identical short placed on random days in the same stocks. The earnings report doesn't make the short trade better. It makes it measurably worse.

Shorting small-cap U.S. stocks after a bad earnings report lost money after costs over the study period, in none of the four time periods examined, and was profitable in only one out of five years.

What is the post-earnings-announcement drift

The finding goes back to academic research from the late 1960s and 1980s (Ball & Brown 1968, Bernard & Thomas 1989): when a company reports its quarterly results, the market only partially digests the news on the day itself. After a positive surprise, the stock tends to keep drifting up for weeks; after a negative one, it tends to keep drifting down. The effect runs counter to the idea of an efficient market, which is exactly why it has been studied for decades.

More recent research (including Chordia et al. 2014, Martineau 2021) finds that in large, widely-traded stocks the drift has largely disappeared — too many market participants watch for it, so it gets arbitraged away. In small, less-followed stocks it's supposed to still be there. This study tests exactly that gap — and only on the short side: selling short after bad earnings.

How this was calculated

Every rule — thresholds, holding period, costs, control groups, and the success criterion — was fixed before the first number was computed (preregistered). Nothing was adjusted after the fact to make the result look better.

The backtest is survivorship-bias-free: stocks that were later delisted (bankruptcy, acquisition, or otherwise) stay in the dataset with their full price history. Dropping failed companies would inflate the apparent performance of any strategy.

An earnings day is defined as the mandatory filing with the U.S. securities regulator in which a company discloses its quarterly results. Because the filing timestamp doesn't reliably indicate before- or after-market release, the reaction day is determined causally: if the stock already breaches the threshold on the filing day itself, that day is the reaction day and the trade enters at the next day's open; if it only breaches the threshold the day after, that day becomes the reaction day instead, with entry one day later still. No price that the trade couldn't have known at entry time ever enters the decision.

Costs include spread and slippage, commission, and a borrow fee of 10% per year — small caps after a sharp drop are often hard to borrow, so a lower rate would have been optimistic. Sensitivity to this assumption is shown further below.

Every number in this article excludes the year 2020. Its extreme price swings would have distorted every average; the year was still simulated, it simply isn't included in the figures shown here.

The main calculation

The core rule: if a U.S. stock with a market capitalization under $2 billion drops at least 10% on its earnings day (or the day after), it is shorted at the next market open and covered 20 trading days later at the close — no stop, because a drift is a slow move and a stop would cut off exactly that slowness.

The median — the typical trade, with half of all outcomes above it and half below — comes in notably milder at −0.61% than the mean at −1.56%. That gap is a feature of shorting itself: a short's loss is unbounded (a stock can in theory rise without limit), while its gain is capped at 100% (a stock can't fall below zero). That asymmetry drags the mean down harder than a handful of strong rebounds should, further than the typical trade actually shows.

The control checks — the real finding

A single negative number doesn't prove the earnings report was to blame — maybe stocks lose money after any sharp drop, regardless of the cause. So two control arms ran alongside the main test: the identical short on random days in the same stocks, and the identical short after a matching price crash with no earnings report.

Against random days, the gap is clear: the short after bad earnings underperformed by −1.15 percentage points (confidence band −1.89 to −0.43 points, which excludes zero). The earnings report is not an edge for the short seller — it's a handicap.

Against the matching-crash control, there is no reliable difference: +0.67 percentage points in favor of the main cell, with a confidence band of −0.64 to +1.94 points — zero sits right in the middle. In plain terms:

The textbook anomaly isn't tradable on the short side.

It isn't the bad earnings report that ruins the short — it's the crash itself. A stock that has already dropped 10% or more in a single day tends toward sharp, hard-to-predict rebounds over the following weeks, whether or not a quarterly number triggered the drop. The classic post-earnings-drift, as described in the literature, provides no independent edge on the short side of small caps.

No arm works

31 preregistered variants were tested in total — different signal definitions, holding periods, size classes, and exit rules. By chance alone, at a 95% confidence level, about 0.8 of these cells would be expected to come out positive. Not a single one of the 31 did.

Signal definitions

Besides the raw price reaction (main cell), the test also ran the academic definition based on the earnings surprise (SUE — standardized unexpected earnings, meaning a quarter's earnings surprise relative to its usual swings over the prior eight quarters), a combination of both, a counter-check, and a blind short on every earnings day.

Holding period

The longer the position stays open, the worse the result gets — a pattern that argues against a durable drift: a real drift should firm up over time, not dissolve.

Size classes

Even within small caps there's no rescuing slice: neither the smallest names (micro caps) nor the somewhat larger ones (small caps) turn positive. Mid and large caps are shown for comparison; they were never the target of this study.

Stops and price targets

A stop could, in theory, have kept individual trades from spiraling out of control. In practice it changes nothing about the overall picture — every variant stays underwater.

Costs and the borrow fee

Shorting costs more than buying: on top of spread, slippage, and commission, there's a borrow fee for the shares being sold short. The main cell assumes 10% per year — a realistic rate for small caps right after a crash, when many market participants want to short at once and shares get harder to locate.

There is no borrow-fee breakeven — a rate at which the strategy would come out flat — for the main cell. Even at an unrealistically low 0.25% per year, the result stays negative at −0.70%. And with every cost stripped out entirely — no spread, no commission, no borrow fee — the winsorized mean still sits at −0.48%. The real problem isn't the cost load; it's the price path itself. On average, these stocks rebound too often and too strongly for a short to outrun them.

Does the finding hold over time

Neither individual years nor broader time periods show a stretch where the rule made money.

Of 15 years measured, only 2011, 2017, and 2022 were profitable — 3 of 15, or 20%. For comparison, a rule that actually worked would be expected to show far more winning years than that. None of the four broader time periods averaged positive either.

What this means — and what it doesn't

The literature's claim that the post-earnings-drift survives in small caps did not hold up on the short side in this study. Whether the same claim fares better on the long side — buying after good earnings — wasn't tested here; the underlying asymmetry of gains and losses is reversed for a long position (gain theoretically unlimited, loss capped at 100%) and would need its own separate test. Other trading approaches on U.S. stocks have already been preregistered and tested, including breakouts following episodic price pivots and several intraday strategies, each with its own independently documented results.

Three limits of this study belong in an honest accounting:

• Borrow availability wasn't actually verified. The calculation assumes every stock was borrowable at the assumed 10% annual rate. Whether a broker would actually have made the shares available — especially for small stocks right after a crash — isn't ruled out, but it isn't proven either.
• The choice of reaction day is a convention, not a fact of nature. Because the filing timestamp doesn't reveal whether a number came out before or after market close, a rule for choosing the reaction day had to be set. Other plausible variants of that rule move the result only modestly (roughly −1.4% to −1.9%) — the overall conclusion doesn't change.
• The earnings-surprise measure (SUE) uses today's data history. The volatility behind this metric is computed from the quarterly figures as they stand today; later restatements of the comparison quarters are baked in, even though the published value itself is used as of its actual release date.

Close

Shorting small-cap U.S. stocks after bad earnings reports lost money after costs over the study period, worked in none of the 31 preregistered variants, and even underperformed the identical short placed on random days. Against a matching price crash with no earnings report, though, there was no measurable difference — the price action following a sharp drop was the real driver, not the earnings report itself.

Source: own backtest of price and fundamental data, 2011 through 2026 (excluding 2020), 5,337 U.S. stock series with 153,715 tradable earnings days. Figures as of August 25, 2026. Design preregistered before the first number was computed.

This article is a historical analysis of publicly available price and fundamental data and is not investment advice. It contains no buy or sell recommendation, no forecast, and no statement about any individual company listed today. Past results — simulated or real — are not a reliable indicator of future returns. Anyone making investment decisions should assess their own situation and risks, and seek professional advice where in doubt.