Refutations

Rob’s framing was not that these tools are unfashionable. It was that they are not true, and that people lose money believing them. He was also careful about who he blamed: “If you make money following these things, you’re not making money for the reasons you think you are. People who follow this stuff and profit are good risk managers — that’s the good part. But the analysis itself is doing nothing.”

A seed does not become a tree until it dies. Each belief below gets crucified, and what grows in its place is a page in these docs.

The standard: what exact means

The claim under every one of these methods is that price obeys a rule — a ratio, an average, a shape, a wave count. If that were true, the rule would be exact, the way applied mathematics is exact everywhere else it governs something.

This is what exact looks like. In October 2020, after a flight of more than 300 million kilometres, NASA’s OSIRIS-REx touched asteroid Bennu within one metre of the spot it had chosen — on a body whose only safe landing area was 16 metres across. And the distance from the Earth to the Moon is known to about one millimetre, by bouncing lasers off reflectors Apollo left on the surface.

Rob put the same point with a calculator: “2 × 2 = 4, always. You don’t get an A for 4.0001.” If price moved by Fibonacci, you could Fibonacci the entire market — retracements and extensions — and know what it was going to do to the penny. Nobody can. That is not a small shortfall in an exact law. It is the absence of one.

1. Fibonacci

The claim: price retraces and extends in the Fibonacci ratios — 23.6, 38.2, 50, 61.8 and 78.6 percent of a move.

The test. Apple, 2024. The swing is real and unambiguous: a low of 196.00 on 5 August to a high of 237.49 on 15 October. Every standard ratio is drawn off it below. Price reaches only two of them, cuts through both without pausing, and turns at 219.71 — in the empty space between the 38.2% and 50% levels, $1.93 below one and $2.97 above the other.

Then the other half of the claim. Fibonacci extensions project the target. From that turn, the 100% extension sat at 261.20. Apple’s highest high over the following months was 260.10 — short by $1.10. The 127.2% missed by $12.39, the 161.8% by $26.74, the 200% by $42.59, the 261.8% by $68.23. One swing, fibbed at both ends exactly as the method instructs, and neither end was found.

And at scale. Across 245 swings in 50 large-cap stocks, the turn landed a median of 0.85% from the nearest Fibonacci level. Five evenly spaced lines that contain no Fibonacci number at all — 20, 35, 50, 65 and 80 percent — scored 0.90%. Indistinguishable. The ratios are not doing the work; five lines drawn across the region where a retracement must bottom are doing the work. And price landed within 0.05% of a ratio in only 5% of cases: 95 times out of 100, the calculator is off by more than a tick.

Crucified. What grows in its place: magnitude is not a ratio, it is a level that already traded — the extreme of a specific prior bar. See types of reversals and the take action window.

2. Moving averages and oscillators

The claim: price bounces off, rides, and finds support at moving averages.

Rob’s objection first. “Lagging indicator” is an oxymoron. “If you come to Chicago and follow me on a walk, and I turn around and ask ‘do you know where I’m going?’ — no, you don’t, because you’re following me.” Price cannot bounce off an average, because bouncing requires mass and price has none. It is not momentum, which is mass × velocity. And it is not self-fulfilling: if it were, everybody would queue at the 200-day and it would never trade again. Somebody has to break price.

The lag, measured. Across 3,582 price turns in 50 large caps, the 50-day average turned a median of 27 sessions after price did, and the 200-day a median of 36 sessions after. It turned first zero times. Not rarely. Never.

The support, measured. Taking every bar whose range contained the 200-day average — 2,807 of them — price was on the other side of the line ten sessions later 50.6% of the time. A coin flip. The same test run against that average shifted 3% away, a line with no meaning whatsoever, reversed 52.7% of the time. The meaningless line scored higher.

Rob’s example. SPY through COVID. In 2020 alone price closed across its own 200-day average ten times — five of them in seven sessions between 27 February and 5 March, alternating down, up, down, up, down. Then it fell to 218.26 on 23 March, finishing 28.1% below the line that was supposed to support it. And here is the lag in one number: the 200-day was 303.37 on 19 February and 303.36 on 23 March. Price fell 35%. The indicator moved one cent.

The oscillators — the RSI, MACD, stochastics. Same claim, same failure, one step further removed: each is arithmetic on past closes, so each can only report what price already did. Rob grouped them with the averages in the same breath: “We can throw in a bunch of oscillators and MACD and RSI — they have no bearing on the future.”

The RSI’s specific promise is a number for overbought and oversold — a reading above 70 or below 30 that is supposed to mean the move is due to stall. Rob’s answer to that promise never changed: “When people say ‘the markets are too overbought, gotta crash’ — you have to have a reversal first.” A trend does not owe anyone a pullback because a formula on its own past closes crossed a threshold; participants who are paying up can keep paying up. Watching a falling stock, he put it the other way round: “Stochastics says oversold? They can always puke more.” And February 2018 supplied the mirror image — the trader who “can’t buy them up here — too overbought on my oscillator” sat out the rise, then had no plan when the actual reversal finally came.

Asked directly whether the RSI can at least identify corrective activity (Power Hour, 27 September 2021), he declined the tool and restated the structure: “People will use all sorts of things. I would say no. You can see if it helps you, but what you’re looking for is a reversal against the direction we’re looking to go, and then another reversal back in our direction.” That is a definition made of price — two reversals — not of any formula built on top of it.

Crucified. What grows in its place: the only line that means anything is an open — the price a group of participants is measured from (see full timeframe continuity). And what the RSI was reaching for is read directly from price instead: overbought and oversold are exhaustion risk — proximity to a broadening-formation extreme — and a pullback is corrective activity, a reversal against the direction followed by a reversal back into it.

3. Pattern recognition

The claim: named shapes — head and shoulders, flags, wedges — predict what comes next.

Rob’s objection. “Pattern recognition is not a Rorschach test. What you think you see is totally irrelevant.” And the books give the game away: the chapter after head and shoulders is failed head and shoulders. “So out of this pattern, price could go up or down? Gee, thanks.”

Measured. Detecting head and shoulders mechanically — head more than 2% above both shoulders, shoulders within 5% of each other, neckline broken on a close — produced 109 completed patterns across 50 large caps. 28.4% reached the textbook measured-move target. 48.6% traded back above the head, which invalidates the pattern completely. Once confirmed, it was destroyed almost twice as often as it worked.

And then it vaporises. Take one of them — Apple, summer 2024, left shoulder in June, head at 237.23 on 15 July, right shoulder on 2 August, neckline broken on 5 August. On the daily the shape is unmistakable. On the weekly it is nine bars. On the monthly it is three. On the quarterly the entire pattern is one candle: 2024 Q3, open 212.09, high 237.23, low 196.00, close 233.00 — green, and closing near its high. Against the previous quarter it made a higher high and a higher low. One timeframe up, the most feared pattern in the textbook is a 2u.

That neckline break on 5 August is the same 196.00 that anchors the Fibonacci leg above. The same forty trading days refute two separate beliefs, which is worth stating plainly rather than leaving to look like coincidence.

Crucified. What grows in its place: three scenarios, only three, and which one you are in is a fact rather than an opinion. See price only trades in 3 ways.

4. Elliott Wave

The claim: price moves in five-wave impulses and three-wave corrections, governed by rules.

Rob was generous about it: Elliott “was on the right track — it understood that range extensions and exhaustions exist.” The problem is that it cannot be quantified, because it is subjective to the analyst.

Measured, using Elliott’s own rules. The three rules that may never be broken are: wave 2 never retraces more than all of wave 1; wave 3 is never the shortest of 1, 3 and 5; wave 4 never enters wave 1’s territory. Take two years of SPY, find its swing pivots, and count every labelling that breaks none of those rules. There are 1,862 of them at a coarse pivot setting and 58,770 at a finer one. The same exercise on Apple gives 105,054.

The rules do not narrow the answer to one. They permit tens of thousands, all equally legal, predicting different things. A method whose rules admit 58,770 valid readings of the same chart is not telling you what price will do; you are.

Crucified. What grows in its place: price discovery as a broadening formation, which is defined by the outside bar and therefore cannot be argued with. See truth #3.

5. Volume

The claim: volume confirms a move. A breakout on heavy volume is real; on light volume it is suspect.

Rob’s objection: the same people say “volume’s too low on the rally — fake rally” and then “volume’s too high — blow-off top.” A claim that explains both outcomes explains neither.

Measured. Every close above a 20-day high across 50 large caps — 4,641 breakouts — split by whether volume was above 130% of its own 20-day average. Ten sessions later, the heavy-volume breakouts were up a median of 0.76% and the unremarkable ones 0.72%. Higher ten days on: 57.1% against 56.3%. The difference in mean follow-through is 0.06 percentage points.

Crucified. What grows in its place: aggression is read from where price is relative to the opens, not from how many shares changed hands. See truth #2.

6. The risk/reward ratio

The claim: “I’ll buy at 100, stop at 99, target 103 — that’s 3:1.”

Rob’s objection: “No it isn’t. Take a statistics class: you have a much higher probability of moving 1 before you move 3. By that logic, why not put your target up 10 bucks with a stop at one cent?”

Measured. Around 10,300 entries per row across 50 large caps, one R defined as one ATR, asking only which came first:

Rob’s claim holds exactly: price reaches 1 nearly twice as often as it reaches 3. And his reductio answers itself — take the 10:1 trade that looks magnificent on paper and you are wrong 89% of the time.

Be precise about what this does and does not show. A 3:1 trade needs a 28.3% hit rate to be better than break-even at 25%, so the numbers above are not an argument that distant targets are bad. They are an argument that the ratio is not evidence. It is two numbers you chose. It says nothing about the probability of reaching either one, and that probability collapses as the target grows. Quoting a ratio describes your intentions, not the market’s.

Crucified. What grows in its place: the target is not a ratio you picked, it is the magnitude of the specific signal you took. See win rate doesn’t matter and measured move.

How these were tested

Every number on this page comes from daily bars for roughly 50 large-capitalisation US stocks, mostly 2021–2026, pulled from a market data vendor rather than hand-picked. The pattern, swing and breakout definitions are stated with each test so they can be disagreed with. Where a control exists it is reported even when it is unflattering: the Fibonacci ratios did beat five randomly drawn ratios, they simply did not beat five evenly spaced ones, and that is the honest version of the result.

The point is not that these tools always fail. It is that they do not do the one thing they claim to do, and that the claim was always testable.

Frequently asked

Does Fibonacci work in trading?

Not as mathematics. Across 245 swings the turn landed a median of 0.85% from the nearest Fibonacci level, and five evenly spaced non-Fibonacci lines scored 0.90% — indistinguishable. Price landed within 0.05% of a ratio in only 5% of cases. If price genuinely moved by these ratios you could map retracements and extensions across the whole market and be right to the penny, the way applied mathematics is right to the millimetre elsewhere.

Do moving averages act as support and resistance?

No. Across 2,807 touches of the 200-day average, price was on the other side of it ten sessions later 50.6% of the time — and the same test on that average shifted 3% away, a meaningless line, gave 52.7%. The average also turned after price on all 3,582 turns measured, and never once before it.

Does the RSI work in trading?

No. The RSI is arithmetic on past closes, so it can only describe what price already did — Rob’s grouping was “oscillators, MACD, RSI: no bearing on the future.” Its overbought and oversold readings are not signals: a move does not stall because a formula reached 70, it stalls when a reversal actually prints. Asked whether the RSI identifies corrective activity, Rob’s answer was “I would say no” — corrective activity is a reversal against your direction followed by a reversal back into it, measured on price directly.

Do chart patterns like head and shoulders work?

Of 109 mechanically detected head-and-shoulders patterns whose neckline broke, 28.4% reached the measured-move target and 48.6% traded back above the head, invalidating the pattern entirely. Any pattern also disappears on a longer timeframe: the Apple pattern of summer 2024 is a single green quarterly candle that made a higher high and a higher low.

Is Elliott Wave objective?

No. Counting every wave labelling of two years of SPY that breaks none of Elliott’s three inviolable rules gives 1,862 valid counts at a coarse pivot setting and 58,770 at a finer one. Rules that permit tens of thousands of contradictory readings of the same chart are not determining the answer.

Related: The 3 Universal Truths · Price Only Trades in 3 Ways · Full Timeframe Continuity · Broadening Formations · Rob Smith
Start here: The 3 Universal Truths →