Volume How Stampede works

What retail volume pressure tells you, and what Stampede does with it

Stocks with a big retail following trade twice as much of their volume in surges as quieter stocks, and their surges keep going. A surge reliably means a bigger next hour. It does not tell you which way. Atty Stampede Algo is built around that split.

Atty Labs · 24 September 2026 · 12 min read

19.8% of crowd-stock volume trades in surge bars, against 9.0% on the control stocks

85% of crowd-stock surges are followed by another busy bar (controls: 64%)

1.17× the next hour’s range after the loudest bars, against normal for that time of day

48% of those next hours moved the same way as the loud bar

Where this came from

We built Stampede to measure how hard volume is pushing price, and which way. This is the long answer to what that means and whether it holds up: where the idea comes from, why crowd-driven stocks are where it should matter most, what it actually predicts when you measure it, and what Stampede does with the reading.

Stampede’s own pressure formula and settings are not published, so the measurements here use public textbook versions of the idea. The one section that tests Stampede itself runs its own code on the same bars and reports what it does, not how it is built. The questions and the pass bars were written down before any of it was run.

The idea is sixty years old

Volume is the footprint of the crowd. A price move on light volume is a few people agreeing; the same move on three times the usual volume is a lot of people acting at once. Joseph Granville turned that into a running tally in 1963 with on-balance volume, which adds a day’s volume when price closes up and subtracts it when price closes down.¹ Alexander Elder’s force index, from 1993, multiplies the change in price by the volume that produced it, so a big move on big volume scores far higher than either one alone.²

Call that reading volume pressure: how hard the crowd is pushing, and in which direction. The part that gets less attention is what the academic record says about it. Jonathan Karpoff’s 1987 survey found that volume rises with the size of a price change in every market studied, and in stock markets runs somewhat higher on up moves than on down moves.³ Volume is tied firmly to how much price moves, and only loosely to which way. That distinction turns out to be the whole story.

Why retail makes the pressure louder

Individual investors do not arrive at a stock one at a time. Barber and Odean showed that they are net buyers of whatever grabs attention: stocks in the news, stocks with unusual volume, stocks that have made an extreme move.⁴ Barber, Odean and Zhu found that their trades are correlated — they tend to buy and sell the same stocks at the same time.⁵ And when Robinhood users crowded into the same names, the stocks they bought most went on to lose about 4.7% relative to the market over the next 20 trading days.⁶

So a stock with a large retail following should trade in bursts: long quiet stretches, then a lot of orders arriving together and pointing the same way. That is what a stampede looks like on a 15-minute chart, and it is what Stampede is named for.

A price chart cannot tell you whose orders made a bar. Separating retail trades from institutional ones needs trade-level data and a specific method.⁷ What a chart can do is compare stocks. So we picked five stocks with a well-known retail following — Tesla, GameStop, AMC, Palantir and Robinhood — and five large, institutionally held stocks with very little of one — Coca-Cola, Procter & Gamble, Johnson & Johnson, PepsiCo and McDonald’s. Both lists were fixed before any number was computed. Each stock contributes 780 full sessions of 15-minute bars from July 2023 to August 2026.

Crowd volume arrives in surges, and the surges keep going

Call a bar a surge when it trades at least three times the usual volume for that 15 minutes of the day, where usual means the median of the previous 20 sessions. On the crowd stocks, 5.9% of bars are surges and they carry 19.8% of all the volume. On the controls it is 2.5% of bars and 9.0% of the volume. The intervals do not come close to overlapping.

The group average hides a split inside it. GameStop trades 40.7% of its volume in surge bars and AMC 31.6%. Tesla trades only 2.6% — the lowest of all ten stocks — because it is so heavily traded all day that three times its usual volume is a rare event. Palantir and Robinhood sit in the same range as the controls.

The cleaner difference is what happens after a surge. On the crowd stocks, a surge bar is followed by another bar at twice the usual volume or more 85% of the time. On the controls, 64%. Four of the five crowd stocks sit between 84% and 93%, Tesla and Palantir included. AMC is the exception at 72%, level with the busiest controls; the controls run from 52% to 71%. When the crowd shows up in these names, it usually stays for a while.

Left: the share of each stock’s volume traded in surge bars. Right: how often a surge is followed by another busy bar. The group averages carry 95% intervals. How much volume arrives in surges differs stock by stock; how long a surge lasts separates the groups more cleanly.

What a surge predicts: how far

To ask what the pressure reading predicts, every bar gets a force-index score — the price change times the volume, scaled so that one stock can be compared with another — and the bars of each stock are ranked into ten equal groups from quietest to loudest. Then two questions are asked about the hour that follows each bar.

The first is how far price travels. After the loudest tenth of bars on the crowd stocks, the next hour’s range is 1.17× normal for that time of day, with an interval of 1.15 to 1.20. On the controls it is 1.10×. The quieter six groups run slightly below normal, at 0.95 to 0.98. Over the next three hours the effect is smaller but still there: 1.10× on the crowd stocks and 1.07× on the controls. Pressure tells you that a move is not finished moving, and it says so more loudly on the crowd names.

Every bar ranked by the size of its force-index reading, quietest on the left. The range of the next hour climbs steeply at the loud end. The direction of the next hour barely moves off 50% anywhere on the scale.

What it does not predict: which way

The second question is whether the next hour moves in the same direction as the loud bar. On the crowd stocks it does 48.1% of the time (interval 47.0% to 49.3%); on the controls, 49.0%. Before running this we set the bar for calling direction informative: the interval has to exclude 50% and the effect has to be at least two points. Neither group clears it at one hour. The average move in the bar’s direction is indistinguishable from zero in both.

At three hours the crowd stocks do clear the bar, and in the unexpected direction: 47.8% of loud bars are followed by a move the other way, a fade of 2.2 points. The controls stay at a coin flip.

Stock by stock, the picture is less tidy than either average. Three of the ten intervals exclude 50%, and they disagree with each other. AMC’s loud bars faded, 43.3% (40.8% to 46.0%), and GameStop’s faded, 46.7%. Tesla’s went the other way: 53.9% of its loudest bars were followed by an hour in the same direction (51.6% to 56.0%). Every other stock is a coin flip. The two meme stocks fading fits the Robinhood finding above: when the crowd piles in together, the price tends to give some of it back.⁶

The loudest tenth of bars in each stock, and how often the next hour followed the push. About 1,700 bars per stock. There is no single direction rule here: Tesla followed, the two meme stocks faded, and the rest sit on 50%.

Why this matters for anyone using a volume indicator

A tool that buys when a big green bar prints on heavy volume, and sells when a big red one does, is trading the part of the reading that carries almost nothing. The loud bar tells you the next hour will be bigger than usual. It does not tell you whether that bigger hour will go your way, and on the most crowd-driven names it leans slightly against you.

How Stampede uses the reading

Atty Stampede Algo is designed around that split. Its pressure reading never picks the direction of a trade. It sets how much room price is given.

Stampede draws a trailing band around price. While pressure is building, the band gives price more room; as the push fades, it closes in again. Direction comes from price alone: the band turns when price closes through it. A turn is not a signal. It becomes one only when a short-term trend and a long-term trend both agree with it, and then the signal is printed at the close of the bar with its stop at the recent swing and its target at three times the risk already drawn. In the published backtests, entries are limit orders at a slightly better price than the signal bar’s close, rather than orders that chase it.

Where the pressure reading goes. It decides how much room price gets and when the band is ready to turn; price and trend decide the direction and whether to act at all.

The real chart below shows it on Tesla over the four sessions in our Stampede bar-replay video. On the open of 19 August the band turned up, both trend checks agreed, and it printed a buy. When a second push arrived at 10:30 the band stayed where it was and let price run; as that push faded over the next hour, it stepped up behind price by about $9.50. The next morning it turned down on the open in a burst of selling, started well above price, closed in as the selling eased, and turned back up before noon without a signal. The lower panel is the pressure reading as Stampede draws it in its own pane.

Tesla, 18 to 21 August 2026, computed by Stampede’s own logic at its default settings. Hollow circles are turns that were not signals. This stretch was chosen because it is the one in our video; it illustrates the mechanism and is not evidence of anything.

What pressure-scaled room buys, and what it costs

Whether giving price room based on pressure does anything useful is testable. We ran Stampede’s band on all ten stocks and compared it with bands that use the same trailing rule but a fixed width that ignores pressure.

The comparison written down in advance used a fixed band with the same average width. Against it, Stampede’s band turns far more often — 1.90 times a session on the crowd stocks against 0.59 — and 16.6% of its turns are undone within the hour, against 3.0%. That is reported as registered, but it mostly shows that a fixed band of that width sits a long way from price nearly all the time. So we added a second comparison after the first run: a fixed band set to turn exactly as often as Stampede’s.

Matched on turns, the differences are smaller and they point in a consistent direction. When a surge lands on the band’s side, Stampede’s band holds through the next hour 85.2% of the time against 82.4% for the fixed band. Its turns are undone within the hour slightly less often, 16.6% against 18.5%. The cost is speed in the other direction: when a surge goes against the band, Stampede’s turns to join it within the hour 64.5% of the time, against 72.1% for the fixed band. The control stocks show the same three differences, and each one has the same sign in at least nine of the ten stocks.

Stampede’s band against two fixed-width bands on the crowd stocks. Turn rate and quick reversals against the same-width band were registered in advance; the surge measures and the same-turn-count band were added after the first run and are exploratory. Lines are 95% intervals.

Put together with the first half of this study, that trade is the sensible one. A surge against the band does not predict the next hour’s direction, so being slow to follow a single counter-surge gives up less than it looks. A surge on the band’s side predicts a bigger hour, so staying with it is where the extra room earns its keep. The band still turns about twice a session, which is why a turn on its own is never a signal.

What to take from this

Read volume pressure as size, not direction. The loudest bars are followed by an hour 10% to 17% bigger than usual. They are not followed by an hour that reliably goes the same way.
Crowd stocks differ more in how long the crowd stays than in how often it arrives. Four of five crowd stocks keep a surge going at least 84% of the time; no control does better than 71%. Surge frequency splits the groups far less cleanly.
Do not carry a direction rule from one crowd stock to another. Tesla’s loud bars followed through; AMC’s and GameStop’s faded. Any rule that assumes one behaviour for “retail names” is wrong for part of the group.
A band that reads pressure trades speed for patience. It holds with a push more often and gets shaken out slightly less, and it is slower to turn against a push. The turn still needs a filter before it is worth acting on.

What we’d test next

Identify retail trades directly. The sub-penny method in reference 7 can flag retail-originated trades in trade-level data. The question is whether surges made mostly of retail trades fade while surges made of institutional trades follow through.
Split surges by news. The meme-stock fade may belong to surges with no news behind them. A split of the same surges into news and no-news is a cheap next test.
Check Tesla’s follow-through on data it was not picked on. It is one stock in ten and the only one on its side of 50%, and a longer history would show whether it is a property of the stock or of these three years.

Method

Data: 15-minute regular-session trade bars from Interactive Brokers, 3 July 2023 to 21 August 2026, for ten stocks. Shortened sessions (eight in the window) are dropped, leaving 780 full sessions and 20,280 bars per stock, 7,800 stock-sessions in all. The first 20 sessions of each stock only seed the volume baseline. The crowd-stock half of the study was first run on a second vendor’s bars and agrees on every measure except surge persistence, which reads 77% on that feed against 85% here.
The two groups: Crowd stocks: TSLA, GME, AMC, PLTR, HOOD, chosen for a well-documented retail following. Controls: KO, PG, JNJ, PEP, MCD, chosen as large institutionally held stocks with little retail attention and no stock split in the window. Walmart was considered and dropped before any data was pulled because of its February 2024 split. The groups are a proxy at the level of the stock; no individual trade is classified as retail.
Measures: Relative volume is a bar’s volume divided by the median volume of the same 15-minute slot over the previous 20 sessions. The force-index score is the change in close from the previous bar times the bar’s volume, divided by the 14-bar average true range at the previous bar and by the same slot-median volume. Bars are ranked into ten equal groups by the absolute score within each stock. The next hour is the next four bars in the same session, measured in the same average-true-range units; its range is divided by the stock’s average for that time of day.
Intervals: 95% intervals from 2,000 bootstrap resamples of whole sessions, using the same drawn sessions for every stock so that market-wide days stay together. Group figures are equal-weight averages of the five stocks.
Pre-registration: The four questions, the group lists, the definitions and the direction bar (interval excluding 50% and an effect of at least two points) were written down before the first run. Two things were added afterwards and are labelled exploratory: the surge join and hold measures, and the fixed band matched on turn count.
The band comparison: Stampede’s band is computed by its own code at the indicator’s default settings, checked bar for bar against the validated engine on 20,366 Tesla bars from a second data feed, with no differences after warm-up. The two comparison bands use the same trailing rule with a width that is a fixed multiple of the average true range, set per stock to match either the pressure band’s average width or its number of turns. Roughly the first eight sessions of each stock are excluded while the bands warm up. A surge here is a bar in the loudest tenth of the force-index score. Joined means the band is on the surge’s side at some point from the surge bar to the end of the next hour; held means a surge on the band’s side is not reversed at any point in the next hour.
What is not published: Stampede’s pressure construction, its lookbacks, its band settings, its trend-filter lengths and its entry offset. Everything the study says about the concept uses the public measures above.
Limits: Three years and ten stocks. No costs, no trades and no profit or loss anywhere in this study; it measures what the reading predicts and what the band does, not whether trading it makes money. The Tesla example is an illustration chosen because it matches the video, not a sample.

Every stock, every headline number

Stock	Volume in surge bars	Next bar also busy	Next-hour range after the loudest tenth	Next hour followed the loudest tenth
Crowd stocks (average)	19.8%	85.3%	1.17×	48.1%47.0% to 49.3%
TSLATesla	2.6%	91.5%	1.15×	53.9%51.6% to 56.0%
GMEGameStop	40.7%	84.1%	1.21×	46.7%44.3% to 48.9%
AMCAMC Entertainment	31.6%	71.6%	1.25×	43.3%40.8% to 46.0%
PLTRPalantir	10.9%	92.7%	1.13×	48.5%46.2% to 50.7%
HOODRobinhood	13.0%	86.8%	1.13×	48.2%46.0% to 50.3%
Control stocks (average)	9.0%	63.6%	1.10×	49.0%47.8% to 50.3%
KOCoca-Cola	6.6%	58.0%	1.11×	50.2%47.6% to 52.7%
PGProcter & Gamble	6.4%	51.5%	1.10×	48.7%46.4% to 51.1%
JNJJohnson & Johnson	12.7%	71.3%	1.11×	49.2%46.7% to 51.5%
PEPPepsiCo	8.7%	66.1%	1.11×	48.7%46.4% to 51.1%
MCDMcDonald's	10.5%	71.0%	1.09×	48.5%46.2% to 50.7%

References

Granville, Joseph E. Granville’s New Key to Stock Market Profits. Englewood Cliffs, NJ: Prentice-Hall, 1963. Link
Elder, Alexander. Trading for a Living: Psychology, Trading Tactics, Money Management. New York: John Wiley & Sons, 1993. Link
Karpoff, Jonathan M. “The Relation between Price Changes and Trading Volume: A Survey.” Journal of Financial and Quantitative Analysis 22, no. 1 (1987): 109–126. Link
Barber, Brad M., and Terrance Odean. “All That Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors.” The Review of Financial Studies 21, no. 2 (2008): 785–818. Link
Barber, Brad M., Terrance Odean, and Ning Zhu. “Do Retail Trades Move Markets?” The Review of Financial Studies 22, no. 1 (2009): 151–186. Link
Barber, Brad M., Xing Huang, Terrance Odean, and Christopher Schwarz. “Attention-Induced Trading and Returns: Evidence from Robinhood Users.” The Journal of Finance 77, no. 6 (2022): 3141–3190. Link
Boehmer, Ekkehart, Charles M. Jones, Xiaoyan Zhang, and Xinran Zhang. “Tracking Retail Investor Activity.” The Journal of Finance 76, no. 5 (2021): 2249–2305. Link

Historical behaviour of ten stocks and of an indicator’s band, not a strategy result or a recommendation. Atty Stampede Algo draws signals, stops and targets on your chart and places no orders. Full disclaimer.