Parabolic Death Cross Study: What the MA10/MA20 Cross Is Actually Worth After a Parabolic Run

Anyone who has watched charts for a while knows the pattern: a stock starts to accelerate, the candles get steeper, the 10-day moving average pulls well above the 20-day average — a parabola. And at some point it tips over: the short average falls below the long one, the "death cross." Many read that as a clear sell signal, some as an invitation to go short.

We wanted to know whether either read holds up, so we ran the numbers on 147,403 such crossings in US stocks since 2010, survivorship-free, with real delistings instead of only the survivors. The result cuts both ways: the cross works as an exit signal — it does not work as a short recipe, because the typical trade wins while the average across all trades loses.

Setup: 147,403 crossings, two applications, one surprising split

The pattern is simple: SMA10 above SMA20 (daily basis, adjusted closes) during a parabolic rise, then the crossing SMA10 below SMA20 — the death cross. Of the 147,403 raw events, 3,396 edge cases fall out (crossing on or after May 2026, too close to the data edge for a complete outcome window). Of the remaining 144,007, the main analysis applies further quality filters — data breaks removed (7.2% of raw events, including one apparent stock that turned out to be two different companies spliced under one ticker), plus suspected splits and price gaps removed, raw price at least $1, median dollar volume at least $1 million — landing at 74,742 events. A narrower subset, called "STRONG," additionally requires a rise of at least 100% AND a high that is also the 120-day high: 3,905 events. Two questions were put to this data: does the cross work as an exit signal for someone who already holds the stock (Application B)? And does it work as a short entry (Application A)? The answers diverge in a surprising direction.

For holders: insurance, not a return source

For STRONG events, the drop after the high is brutal: the median peak-to-trough loss is 48.5%. What matters for whether the cross works as a warning: more than half of that drop — 51.5% at the median — happens AFTER the cross, not before. Selling at the cross instead of holding does not miss the main move; it avoids a substantial part of it.

At first glance, a clear yes: selling at the cross beats holding, on the median and in the majority of cases. But a closer look tempers that considerably: the winsorized mean of the savings — a robust average across ALL cases that caps extreme outliers rather than ignoring them — sits at -2.2 percentage points at 60 days. Slightly negative. The honest reading: the cross signal is insurance against catastrophe, not a source of extra return. It reliably caps the worst crashes, but costs something on average, because it occasionally sells a stock that goes on to run again — a continuation a holder would have missed. For moderate rises (30-100% instead of the full STRONG definition), the effect nearly disappears: there, the median savings in the main analysis sits near zero or slightly negative. The signal works specifically for genuinely strong parabolas, not for every MA10/MA20 cross.

For short traders: wins on the median, loses on average

This is where the real point of this study begins. A short entered at the cross looks tempting at first glance:

The median and hit rate favor the short: more than half of all trades win, and the typical gain runs 6.6%. The winsorized mean flips negative, though. The cause lies in the SHAPE of the distribution, not its center: most short trades produce moderate gains, but a small minority runs extremely against the short — a squeeze, where the price explodes instead of continuing to fall. One example in the underlying data, from 2026, shows a position that ran +226% against the short. Rare but extreme outliers like that arithmetically eat back the gains of many small winners. The median shows what the typical trade does; the mean shows what a portfolio of very many such trades would have earned on average. For this signal, the two answers point in different directions.

In the broader main analysis (all rises from 30% up, not just STRONG), the short is completely dead — there, even the median is negative. Only at extreme rises above 400% does an exception appear: there, the winsorized mean turns slightly positive too (+8.5% at 163 events over 60 days) — a rare edge observation with a small sample, not a workable recipe. And none of these figures include borrow costs or actual short-sale availability — both would likely make the picture worse still.

Stops make it worse, not better

An obvious reflex: if rare squeezes wreck the mean, shouldn't a stop-loss help? The opposite is true. Tested were fixed stops at 10%, 15%, 20%, and 25%, a 15% trailing stop, and a re-cross exit (exit as soon as SMA10 climbs back above SMA20, capped at 120 trading days).

Every stop variant massively worsens the median compared to the no-stop version — at stop-out rates between 47% and 70% of all trades depending on the stop distance, and 97% for the trailing stop. A tight stop on a volatile short squeeze almost always triggers first, before the underlying move actually plays out — the stop doesn't protect against the squeeze, it realizes it. The same pattern showed up in our Trio Short study: tight stops aren't risk management for this signal type, they're an extra drag on returns.

Profit targets: better optics, same underlying problem

What if, instead of capping a loss, you lock in a gain? Tested were fixed profit targets of 5%, 10%, 15%, and 20%, otherwise a forced exit after at most 60 trading days. In the table, the median is net (after 0.4% round-trip cost); hit rate and winsorized mean are gross (before costs).

At first glance this looks like the fix: hit rates up to 84%, clearly positive medians. But the winsorized mean stays slightly negative in every variant. A profit target caps gains from above — the maximum payoff on any trade is limited to 5-20% — while a short squeeze running against the position stays uncapped. The result: many small, reliable wins and rare, ruinous losses. The exact pattern that makes the short fail as a recipe persists — it's just better disguised.

Entering earlier? Counterproductive

One more obvious idea: why wait for the cross if the rise is visibly already tipping over? Tested was the first close that lands 5%, 10%, or 15% below the 20-day line (SMA20), searched in the window from the high through 30 trading days after the cross — despite the name, a lagging rather than a leading criterion: at the 5% threshold, this entry lands at a median 2 days BEFORE the cross; at 10%, at a median right ON the crossing day itself (only 48.3% of cases are earlier at all); at 15%, at a median 2 days AFTER the cross (only 31.8% earlier). The result still runs clearly the other way: at the 10% and 15% thresholds, this entry lands, at the median, 3-8% BELOW the later crossing price. At 10%, it loses against the cross entry on an exactly identical subset of events, on every single metric (no such direct comparison exists for 15%). At the 5% threshold, the median difference is practically zero. Entering too early buys (or shorts) into a move that isn't over yet, instead of waiting for confirmation from the cross.

A five-filter test for "genuine" parabolas

Not every STRONG crossing looks like a genuine parabola — some arise from a single one-day jump, some from a slow grinding trend, some from an MA20 break in the middle of the rise. From nine explicitly flagged unwanted patterns, five filters were derived: the share of a single one-day jump in the total rise, the age of the signal since the high (more than 15 days counts as "old"), a slow-grind trend score (persistence above 45), an MA20 break during the rise, and too small a distance of the crossing from the MA20 (below 85%). Applied to the 3,905 STRONG events and combined with the rule "short at the cross, 15% target, 15% stop, day-10 time exit," 842 events remain (839 with a fully computable outcome).

The result is sobering: the filters make the short slightly WORSE, not better — both hit rate and median dip somewhat versus the unfiltered comparison. Hunting for the "genuine" parabola doesn't help the short recipe, because the real problem isn't a few ugly individual price paths — it's the underlying shape of the distribution: rare, extreme counter-moves that no characteristic filter reliably screens out in advance.

Where the signal lives: illiquid names, real information only for STRONG

The effect concentrates clearly in less liquid stocks. At $1-5 million in daily dollar volume, short performance over 60 days is at its best (median slightly positive); already in the next class ($5-25 million) the median turns negative and keeps falling as liquidity rises — worst above $100 million in daily turnover. An intuitive pattern, since large, heavily traded stocks run into extreme parabolic moves less often and are harder to short without triggering squeezes of their own.

To check whether the signal carries real information rather than pure noise, a control group was used: for each event, a random trading day in the same stock, with the same windows. The difference is clear, but only for STRONG events: the real crossing day beats the random day by 4.8 percentage points at the median and 5.0 percentage points in hit rate (60 trading days). For the broader main analysis, that edge shrinks to 0.4 and 1.2 percentage points respectively. So the signal does carry real information — but mostly for genuinely strong, clearly identifiable parabolas, not for any arbitrary MA10/MA20 cross. Excluding the two major market-stress windows (February-April 2020, all of 2022) barely changes the picture — this isn't purely a crash effect.

Limitations of this study

• Borrow costs and availability are entirely absent. None of the short metrics in this study account for the cost of borrowing the stock, or whether it was actually available to short at that point in time — for strongly parabolic, often smaller stocks, both are a real issue and would likely make the result worse still.
• Stops are computed on a closing-price basis. Signal and exit occur on the same day's close; a stop triggered intraday would likely be less favorable in practice, since intraday swings are often more extreme than close-to-close moves.
• Fine-grained subgroups thin out fast. Breaking down by gap size, liquidity class, or filter combination sometimes cuts the sample sharply (e.g. only 163 events for rises above 400%) — such edge cases are more a hint than a robust result.
• Trading costs are a flat assumption. 0.2% per leg approximates bid-ask spread and fees only roughly; for the less liquid names where the signal is strongest, the real drag could run higher.

Anyone who wants to apply this rule to today's candidates can use our new scanner, Parabolic Death Cross — it surfaces stocks that currently form a parabolic death cross under the definition tested here. This isn't a buy or sell signal; it's a tool applying the same criteria examined in this study. Readers who want to see the same survivorship-free approach applied elsewhere can find it in our Trio Short study — where stops hurt returns in the same pattern. More backtest studies live together in Studies.

Figures as of August 5, 2026.

This article is a historical analysis of publicly available price data and not investment advice. It contains no buy or sell recommendation, no forecast, and no statement about any individual company trading today — the accompanying scanner is no substitute for independent research either. Past results — whether simulated or real — are not a reliable indicator of future returns. Anyone making investment decisions should assess their own situation and risks, if in doubt with professional advice.