On September 13, a quarterly return table attributed to CoinGlass showed Ethereum up 60.62% for the third quarter of 2026 — the second-best Q3 on record, behind 2025's 66.55% and just ahead of 2020's 59.5%. The number is correct. The candle is honest. The window is not.
Here is the arithmetic the headline removed. ETH closed Q1 2026 down 29.26%. It closed Q2 down 25.28%. Compound three quarters and you get 0.7074 × 0.7472 × 1.6062 = 0.849. Year-to-date, through September 13, Ethereum sits roughly 15% below where it opened January. The second-best third quarter in the asset's recorded history is resting on a year that is still red.
That is not a caveat. That is the finding, and it was left out of the table.
Context: a note that exists because the number is large
CoinGlass aggregates derivatives and spot market data — liquidation heatmaps, open interest, funding rates, and cross-venue return series. A quarterly return flash note is a low-cost product. Pull the quarterly close, rank it against history, publish. Production cost is near zero and distribution is wide, and that economics shapes what gets made.
The dataset the note cites against itself tells you this. Historical mean Q3 return: 12.28%. Median: 9.87%. Both are good quarters. Neither generates a headline. Nobody publishes a note titled 'ETH Delivers a Slightly Above-Average Quarter.' The note exists because 60.62% cleared a threshold that roughly one quarter in a decade clears. Its existence is a function of the number being an outlier, not of the number being informative.
I learned the gap between an anomalous number and a meaningful number in late 2017, on a contract review for an ERC-20 launch called EtherGem. I found three arithmetic overflow vulnerabilities in the token's voting mechanism using a short Python script — nothing sophisticated, just integer wraparound in a function nobody had tested. I sent the findings to the development team. The token rose 400% over the following weeks and the findings were not acknowledged. Three months later the project collapsed through a rug pull that exploited exactly the overflow paths I had flagged.
The lesson was not that I was right. The lesson was that a price series and a contract state are two different objects, and a rising price series will happily run on top of a broken contract. Price is the last thing to know. It is the first thing to be quoted.
That is the structure of this Q3 note. The price series moved. Nothing was said about contract state.
Core: the base, the distribution, and the missing mechanism
The base effect is not a footnote — it is the load-bearing wall. A 60.62% gain measured from a denominator that lost roughly 47% across the prior two quarters is not the same object as a 60.62% gain measured from a flat base. Drawdown math is asymmetric, and most readers discount it correctly in theory and incorrectly in practice. A 50% loss requires a 100% gain to recover, and this is not a rhetorical flourish; it is division. Two consecutive quarters of -29.26% and -25.28% leave you at 0.5274 of the starting level. A 60.62% recovery takes you to 0.849. You are still down 15%. The recovery is real. The recovery is also incomplete, and the headline reports the recovery without the level.
I built a SQL dashboard in 2020 to run precisely this decomposition on Aave v1's liquidity mining incentives, tracking published APY against actual treasury reserves. The APYs were real numbers. The reserves backing them were not sufficient to sustain them. The published figure and the sustaining mechanism were measured in different units, and everyone quoted the first one. That is the same defect here. The published figure is a quarterly delta. The sustaining mechanism is a year-to-date level. Different units. One got quoted.
The distribution says tail, and tails are not schedules. The note supplies its own refutation and does not notice. Mean Q3 return: 12.28%. Median: 9.87%. Median below mean means the Q3 return series is right-skewed — the average is dragged upward by a small number of blowout quarters while the typical quarter lands below it. In a right-skewed series, the modal outcome is worse than the advertised average. A reader anchoring on 12.28% as normal is already overestimating the normal quarter.
60.62% is roughly five times the mean and six times the median. In a right-skewed quarterly series that is a far-tail observation, and far-tail observations have a specific property: they are descriptive, not predictive. The print is real. The base is poisoned. The distribution is not a trend. The correct inference from a tail print is that something happened, not that something is happening.
Two consecutive record Q3s is the anomaly inside the anomaly. 2025 printed 66.55%. 2026 printed 60.62%. The top two third quarters in the recorded series are adjacent. The note presents this as confirmation of momentum. Under a naive independence assumption it is a far stronger claim than that — two consecutive draws landing in the top two slots is not what independence looks like. Three explanations compete and the note offers none of them.
First, genuine seasonality: mid-year institutional rebalancing, fiscal calendar effects, tax-related flow timing in major jurisdictions. If a repeatable Q3 bid exists, the historical Q3 mean of 12.28% is a contaminated benchmark. Second, structural change in the holder base: spot ETF vehicles created an external, calendar-driven bid mechanism that did not exist in the 2017 to 2021 sample. If the marginal holder changed, the historical distribution is stale, and bounding a structurally new market with pre-ETF statistics is a category error. Third, coincidence. Two observations cannot distinguish these three. The note treats the ambiguity as strength. It is ambiguity.
The dataset has a publisher, and the publisher has a P&L. CoinGlass monetizes market activity. Its product surface — liquidation screens, open interest, funding-rate data — becomes more valuable when markets move violently. That does not imply fabrication. It implies selection. Which slice of a return series gets packaged and pushed is not a neutral editorial act; it is an act performed by an entity whose revenue correlates with the drama of the slice.
I ran into this in 2021 while tracing Bored Ape floor price volatility. Roughly 15% of weekly volume clustered into wallet groups traceable to a single governance-adjacent address. The volume was real on-chain. The volume was not organic. The distinction between a metric being accurate and a metric being meaningful is the entire job, and I turned it into a recurring Wash Trading Index precisely because the industry kept quoting the first measurement and never asked the second question. The same discipline applies here. Return magnitude and return significance are different objects. CoinGlass published the magnitude. Nobody has published the significance.
The missing variables are not peripheral. They are the mechanism. What the note omits is a list of everything that would let a reader convert a delta into a judgment.
Spot ETH ETF net flows across the quarter. This distinguishes an external bid — new capital entering the asset class — from an internal one, where existing crypto capital rotated out of other tokens or borrowed against leverage. These produce identical price charts and completely different forward distributions.
Staking ratio and validator entry and exit queues. Rising stake locks supply and mechanically tightens float. Falling stake releases it. A 60% quarter with rising stake is a supply story. The same quarter with falling stake is a demand story that borrowed against future supply. The note reports neither.
EIP-1559 burn volume. This is the closest thing Ethereum has to a revenue line. Burn rising with price implies fee demand is real. Burn flat or falling while price rises implies the price has decoupled from network usage. In the post-Dencun regime, where blob space pushed L2 costs down and L1 fee revenue down with them, this distinction is not academic. It is the difference between an asset repricing on usage and an asset repricing on liquidity.
Exchange net flows. Coins moving onto venues precede sell pressure. Coins moving off precede accumulation. It is arguably the single most useful on-chain series for reading the sustainability of a quarterly print, and it is absent.
Perpetual funding rates and open interest. A 60% quarter carried by spot bid and a 60% quarter carried by leveraged longs produce the same quarterly return and opposite risk profiles. The second unwinds violently. Without funding and open interest, the note cannot tell you which one occurred, and the reader will default to the benign interpretation.
BTC's Q3 return. This is the most important omission because it is the cheapest to obtain. If BTC printed +45%, then ETH's +60.62% is beta plus a modest rotation premium and requires no Ethereum-specific explanation at all. If BTC printed +5%, then ETH's move is idiosyncratic and demands a specific cause the note does not supply. Code compiles, but context reveals the exploit. Without the cross-asset comparison, 60.62% is a number without a category.
There is a second layer the note cannot reach even in principle. Price is not network state. ETH's price is a claim on the L1 monetary asset. It is not a claim on L2 throughput, L2 fee capture, or L1 settlement revenue. Those are separate ledgers, and they can diverge for years before anyone notices.
Dozens of L2s now compete for a user base that has not grown proportionally. That is not scaling; it is a fixed pool of liquidity sliced into progressively thinner fragments, each with its own bridge, its own sequencer, its own incentive program, and its own TVL to defend. Dencun made L2 blockspace cheap by design, which was correct engineering and simultaneously reduced the fee base flowing back to L1. A rising ETH price and a compressing L1 fee base can coexist for a long time. Nothing in a quarterly return table can see the difference. The note measures the asset. It says nothing about the network.
ETH is also the dominant collateral asset in on-chain credit, and this creates reflexivity worth stating plainly. A 60% quarterly rise mechanically improves the health ratio of every ETH-collateralized loan book, suppresses liquidations, removes forced selling, and thereby supports price further. That loop is real and it is a legitimate bullish mechanism. It is also symmetric. The same loop runs in reverse with equal mechanical efficiency.
In 2022 I compared Frax's partial collateralization to Terra's algorithmic model and concluded both rested substantially on market confidence rather than hard assets — different ratios, same dependency. The collateral layer here has the same property at one remove. A loan book marking ETH collateral at a price that has just traveled five times the historical quarterly mean is not marking against stability. It is marking against an extreme observation in its own historical series. That is not a solvency argument. It is a latency argument: the collateral marks are correct today, and they were also correct at the top of Q2 2022.
Ranking narratives are structurally late by construction. You cannot print a top-two quarterly return until most of the return has already been delivered. The ranking is arithmetically guaranteed to appear near maximum realized gain and minimum remaining visibility. This is not a market-timing claim about what happens next. It is a claim about information content: a ranking is a lagging statistic that reads like a leading one. Every time the industry quotes best quarter since, it is quoting a rear-view mirror and calling it a windshield.
Contrarian: what the bears get wrong
The bearish read — this is a low-base bounce off a flushed market, ignore it — has its own exploit, and it is worth naming because the lazy version of the bear case is as analytically empty as the note.
A low-base bounce is still a bounce, and mechanical conditions matter. If Q1 and Q2 cleared leverage and exhausted the marginal seller, the setup for a sustained repricing is precisely a compressed base plus exhausted supply. That is not an illusion. That is the condition from which trends begin. Dismissing a move because its base was low is a description, not a refutation.
Mean reversion is also a forecast, not a law. The bearish argument that 60.62% must revert toward the 9.87% median assumes the distribution is stationary. In a right-skewed series, the right tail occasionally produces consecutive outliers during trending regimes — that is what a trending regime is. Using the historical median as a hard attractor is the same error as extrapolating the tail: both assume the sample describes the present.
And the standard critique that there is no fundamental attribution cuts in both directions. Ethereum has no cash flow statement. There is no earnings call, no revenue guidance, no filing. Absence of an attribution table is not evidence of absence of a cause; it is evidence that the asset does not produce the artifacts the questioner keeps asking for. Spot ETF access in 2024 and 2025 created a genuinely new external bid mechanism with its own calendar and its own counterparties. If that mechanism has matured, the pre-ETF Q3 distribution is stale, and bounding a structurally changed market with stale statistics is the same error as bounding a 2025 protocol with a 2017 whitepaper.
What the bulls actually get wrong is narrower and more serious than direction. They are treating a single tail print with zero flow data as confirmation of a thesis. A tail print with no flow data is a tail print with no flow data. It is not a thesis, and it is not evidence against one either. It is an unlabeled observation. The note's failure is not that it is bullish. It is that it is unlabeled.
Takeaway
The question is not whether 60.62% happened. It happened, and it is accurate, and it is nearly meaningless in isolation. The question is whether anyone holding ETH on the strength of that headline has computed their own year-to-date number. The plumbing is straightforward — Q1 down 29.26%, Q2 down 25.28%, Q3 up 60.62%, net roughly -15% — and the distance between what the headline implies and what the arithmetic shows is the entire information content of this quarter.
What would change my read is not another return table. It is ETF flow data, staking ratio, burn volume, exchange netflows, funding and open interest, and BTC's Q3 print, read together. When those arrive, the number will have a category. Until then it has a headline, and the headline was selected by a publisher whose revenue rises with the magnitude of the number it is publishing.
The window closes September 30. Whatever the final print turns out to be, the statistic that circulated on September 13 was chosen, not discovered — and the denominator it omitted is the only part of the story that is still open.
