Four for four is not a track record. It is a sample size.
The claim circulating this week โ attributed to analyst Ali Martinez, relayed through X, and repackaged by crypto media โ is that bitcoin has fallen ahead of each of the last four US midterm elections, and that the next one, scheduled for November 3, may trigger another pullback. The four figures attached to the claim are 72%, 65%, 52%, and 27%. Bitcoin is quoted at $73,000 in the same breath.
Nine years of reading claims like this have taught me one procedural habit: read the metadata before the argument. The metadata here is thin. There is no publication timestamp. There is no link to the original post. There is no year in the headline. Three absences, and one of them is fatal.
Silence is the first red flag. Not because the analyst is lying โ I have no evidence of that, and I am not in the business of assigning motive where a structural defect will do. The problem is narrower and worse: a claim that cannot be dated cannot be tested. An undated price forecast is not a forecast. It is a horoscope with a candlestick chart stapled to it.
What follows is not a rebuttal of a bearish view. I hold no position on whether bitcoin falls in the fourth quarter. What follows is a stress test of the machinery behind the claim โ the sampling, the base rate, the time window, the indicator being cited, and the incentives that shaped how it reached you.
The Missing Year
The number 3 in "November 3" does almost all of the work here, and almost nobody noticed.
US federal elections are held on the Tuesday following the first Monday in November. That is statute, not convention, and it produces a different calendar date every cycle. 2010: November 2. 2014: November 4. 2018: November 6. 2022: November 8. 2026: November 3.
Exactly one of those matches the date in the report. That gives a hard constraint: if the source is describing a future event, it is describing 2026. A second constraint stacks on top of it โ the report lists 2022 as a completed cycle with a 27% drawdown, so the document cannot predate November 2022. Cross the two and the artifact is almost certainly a 2026 document, written in the weeks before the vote.
Which means the $73,000 is not a historical snapshot. It is a live quote. And that single fact rewrites the meaning of the entire piece.
If the 2025 high printed above $120,000 โ a figure I would verify against exchange data before publishing anything, and which readers should treat as provisional until they do โ then $73,000 represents a drawdown north of 39% from the high. The market has already corrected. Substantially. The thesis being sold to you is not "a pullback is coming." It is "a second pullback is coming, after the first one already happened." Those are different products, and only one of them is what the headline implies.
Now run the alternative. Suppose the year was stripped in reprint โ a very common failure mode when a post travels from X to a content farm to an aggregator to you. Then $73,000 is a stale print from some unknown date, and every conclusion built on it is void. Not weakened. Void.
Either way, the reader cannot distinguish between a live forecast and a dead one, because the document does not contain the information required to tell them apart. That is the defect. Everything else is downstream of it.
There is a third layer, and it is the one that irritates me most. The headline reads that the midterms may trigger a pullback. The body records the analyst explicitly stating that historical correlation does not mean the election causes the decline, and that price is still governed by liquidity and macro conditions. The headline's causal strength is materially higher than the analyst's own language. That is framing, not reporting, and it is the oldest trick in the trade: put the causation in the title where the clicks are, put the disclaimer in paragraph nine where the lawyers are.
Four Numbers and a Denominator
Here is where the argument actually breaks, and it breaks on arithmetic, not opinion.
The claim rests on four observations. Bitcoin's maximum intra-year drawdown in the four midterm years was, in order, 72%, 65%, 52%, and 27%. All four are large. Therefore midterm years are dangerous.
Missing from that sentence is a denominator. What is the unconditional probability that bitcoin posts a drawdown of 27% or more in any given year? Without that number, "four out of four" is not evidence. It is a numerator looking for context.
So let me supply one. Counting year by year from 2010, the years that produced an intra-year peak-to-trough decline in excess of 27% include 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022, and 2024. The conspicuous exception is 2023, which bottomed out around 20%. Depending on how you score the marginal years โ 2012 and 2016 sit near the boundary โ the base rate lands somewhere between 75% and 90%, and if you count generously it clears 90%.
History is just data waiting to be read, and this is what the data says: bitcoin draws down more than 27% in the overwhelming majority of years, regardless of what is on the ballot.
Now do the probability. If the unconditional rate is 80%, the chance of seeing four hits in four tries by pure chance is 0.8โด, which is 41%. If the unconditional rate is 90%, it is 0.9โด, which is 66%. In plain terms: a coin weighted the way bitcoin already is would have produced this exact "pattern" between forty and sixty-six percent of the time.
The midterm grouping is not a signal riding on top of noise. It is noise wearing a name tag.
This is the point where a Bayesian framing is more honest than a frequentist one. Start with a prior belief that midterm years behave like any other year. Observe four out of four. Compute the likelihood ratio: how much more likely is this observation under "midterms matter" than under "midterms don't matter"? Given a base rate of 80%, the answer is a factor of roughly 1.25. That is a rounding error. Your posterior should barely move. The claim is not wrong in the sense of being false โ it is wrong in the sense of being uninformative, which in a market context is the same thing.
I learned this lesson the hard way in 2017, reverse-engineering a preliminary token distribution schedule in Python while everyone around me was arguing about a whitepaper's adjectives. The document said "decentralized." The allocation table said sixty percent to insiders. One of those was a claim and the other was arithmetic, and the arithmetic won. The same move applies here. The narrative says "midterm cycle." The base rate says "every year." The base rate wins.
The Calendar Is Not a Mechanism
Even granting the sample, the timing does not survive contact with the actual price history.
The cited figures are annual maximum drawdowns โ the largest peak-to-trough move within a calendar year. That is a twelve-month statistic. The midterm election is a one-day event with a roughly two-week information window around it. Comparing them is a time-scale mismatch of the first order, and it is the kind of mismatch that looks like analysis until you write the dates down.
Write them down. The 2014 drawdown was concentrated in February, when Mt. Gox collapsed. The 2018 damage came in two waves โ January through February, and then November, when the Bitcoin Cash hash war split the network's hashrate and the price broke support. The 2022 collapse was May through June, driven by Terra and then Three Arrows Capital, followed by a second leg in November when FTX failed. The 2010 figure belongs to a market that barely existed.
Note what happened in 2018 and 2022: both had November drawdowns, and both Novembers happened to contain an election. It is tempting to call that confirmation. It is not. In 2018 the proximate cause was a mining-hash civil war between two client implementations. In 2022 the proximate cause was the insolvency of the second-largest offshore exchange. Both events were idiosyncratic, both were internal to the asset class, and neither had a causal channel running from a congressional ballot to a bitcoin order book.
A scheduled event carries no information. Only its deviation from expectation does. Every market participant knew the midterm date years in advance. There is no surprise in a calendar. What moves price is the polling error, the policy discontinuity, the unexpected committee chair โ not the Tuesday.
If you want the election to matter, you need to name the transmission mechanism. Which candidates, which committee majorities, which specific legislative outcomes change the marginal buyer's behavior in a way the options market has not already priced? Absent that, you are not forecasting. You are curve-fitting a calendar to a candlestick and calling the residue a signal.
What $73,000 Actually Is
The technical core of the claim is that $73,000 represents the short-term holder cost basis, and that this level acts as support. The first half of that sentence is correct. The second half is where the methodology fails.
Short-term holder cost basis โ standard Glassnode territory โ is the weighted average acquisition price of coins that have moved within the last 155 days. It is a real, computable, verifiable number. I use it. But look at what it measures: the average break-even point of recent buyers. That is a behavioral reference, not a liquidity reference.
The distinction matters mechanically. Order-book support works because resting bids absorb sell pressure โ depth, size, and price level determine how much volume it takes to break through. Cost-basis support works through loss aversion: holders who are underwater tend not to sell, which reduces supply. Those are different mechanisms operating on different variables. Conflating them leads you to overestimate the strength of a bounce, because you are assuming a wall of bids where there is only a wall of reluctance.
Worse, the number is not a floor. It is a rolling value. Coins age out of the 155-day window every single day, and new coins enter it. The weighted average drifts continuously as the composition of the cohort changes. A level computed in August is not the level in October. Anyone treating a rolling average as a static price floor has confused a speedometer for a parking brake.
The honest reading is that the STH cost basis is a two-way magnet. Below it, recent buyers are underwater and supply thins โ mildly constructive. Above it, the same cohort sells into break-even to escape โ mildly destructive. It is a sentiment gauge with a short half-life, and it should be re-read weekly, not cited once in a headline.
The ledger lies; the code tells. The on-chain number is the code. The word "support" is the ledger entry someone wrote next to it.
I ran into the same category error in 2020, modeling Compound's liquidation thresholds under volatility regimes that the protocol's own parameters had never been tested against. The health factors looked fine in the dashboard. They looked very different in a sandbox where the collateral fell 40% in six hours. The lesson stuck: a risk parameter is only meaningful under the stress conditions it was designed for, and a support level is only meaningful at the timescale it was computed on. $73,000 was computed on a 155-day window. It has no opinion about November.
The Trend Line Nobody Quoted
Here is the part of the data set the analyst had in hand and did not put in the post.
72%, then 65%, then 52%, then 27%. The sequence is not random. It is monotonically decreasing, and the decrements are accelerating: minus 7, minus 13, minus 25.
Extrapolate naively โ a straight line through four points โ and the next midterm year lands somewhere near 12%. Extrapolate the acceleration and the next figure goes negative, which is to say, no drawdown at all. Both methods point below 27%. Neither supports a bearish conclusion. The analyst's own sample, read as a series rather than a set, argues against the thesis he is advancing.
He left it out because a decaying bear case does not travel. "Midterms may trigger a pullback, though the effect has shrunk by two-thirds since 2010" is a true sentence and a dead post. I understand the incentive. I am describing it, not excusing it. Incentives align, or they break, and here the incentive to publish aligned with the incentive to truncate.
There is a second, subtler problem with the percentages, and it is the one that gets past almost everyone. Percentage drawdowns compress mechanically as an asset's market capitalization grows. A 72% decline in 2010, when bitcoin's entire network value was under a hundred million dollars, destroyed almost no capital in absolute terms. A 27% decline at $73,000 destroys more dollars than every drawdown before 2017 combined. Percentage compression is not risk compression. If anything, the dollar-denominated volatility of this asset has increased as it has matured, because the pool of capital at risk has increased faster than the percentage has fallen.
Anyone citing the decay trend as evidence that bitcoin is "maturing into a lower-volatility asset" is measuring the wrong variable. Measure the dollars. The line goes the other way.
The Sample That Should Not Be There
The 2010 observation is not a data point. It is an artifact, and including it is a methodological error serious enough to invalidate the average.
In 2010, bitcoin's market capitalization was under one hundred million dollars. There was no derivatives market โ no futures, no perpetuals, no options. There were no stablecoins, because there was no Tether. Mt. Gox had only just launched in July of that year. The participant base was a few thousand cypherpunks and hobbyists on a forum. Liquidity depth was measured in thousands of dollars per side.
Now compare that to a market with a spot ETF complex, a deep options surface, a perpetual swap market running tens of billions in daily volume, and institutional market makers quoting two-sided on regulated venues. These are not the same asset in any economically meaningful sense. They are not even the same market structure.
Stacking 2010 next to 2022 and calling the result an average is like averaging the stopping distance of a Model T and a modern sedan and calling it "the braking performance of cars." The number exists. It means nothing.
Strip 2010 and the sample is n=3. Three observations. No significance test, no confidence interval, no benchmark comparison. Three data points is not a cycle. It is an anecdote with a spreadsheet.
I did this kind of clustering work in 2021, mapping wallet graphs on a marketplace to identify fifteen interconnected addresses wash-trading a blue-chip collection and inflating the floor by roughly two million dollars. The tell was never the volume. Volume is noise; intent is signal. The tell was that the volume existed only between addresses that shared funding sources. Here the tell is the same shape: the sample looks large until you cluster it, and then it collapses to three.
The Hedge
Read the analyst's full language and something structural appears. He flags a downside risk. He also flags that observers should watch whether sell pressure fades and whether buyers step in โ which is a polite way of describing a bottom.
That construction cannot be falsified. If price falls, the drawdown thesis is confirmed. If price rises, the bottom call is confirmed. There is no state of the world in which the analysis is wrong, which means there is no state of the world in which it is right either.
A real risk memo โ the kind I write for clients โ contains four things this does not: an invalidation level, a time window, a position-sizing implication, and a stated probability. Without those, a forecast is unfalsifiable, and an unfalsifiable forecast is not analysis. It is brand maintenance.
Algorithmic truth requires no defense. If you can write the condition that would prove you wrong, you have done the work. If you cannot, you have done something else.
What the Bulls Got Right
I have spent most of this piece dismantling the bear case, so let me be precise about what the other side gets correct โ because the strongest argument here is not that bitcoin will rise. It is that the analyst's own caveat is the actual model.
He says price is governed by liquidity and macro. He is right, and that sentence contains the real thesis. The midterm calendar is a proxy for macro risk appetite. What actually transmits is dollar liquidity, real rates, and credit conditions. When those tighten, risk assets fall. When they loosen, they rise. The election matters only insofar as it changes the expected path of fiscal policy โ and even then, only the surprise component of that change.
And here is the structural point the cycle model cannot accommodate: the marginal buyer changed in 2024 and never changed back. When spot ETFs launched, a new class of price-insensitive, calendar-indifferent, allocation-driven capital entered the market. That capital rebalances quarterly. It does not care about November. It cares about the creation and redemption mechanism, the basis trade, and the model portfolio weight its consultant handed it.
Friction reveals the true structure. The drawdown decay from 72% to 27% is not a political phenomenon. It is the fingerprint of a market maturing โ deeper books, hedging instruments, an options surface that lets holders buy protection instead of selling spot. That is a real, measurable, structural change, and it explains the trend line better than any election cycle ever could.
Which brings me to the risk nobody is pricing. In 2024 I went through the custody structures of the major ETF issuers and found that the overwhelming majority of underlying assets sat in cold storage controlled by third-party custodians under arrangements that would make a self-custody advocate reach for a drink. That is the actual systemic exposure in this market โ not a Tuesday in November. If you want to stress-test bitcoin, stress-test the custody chain, the redemption plumbing, and the concentration of authorized participants. Those are the joints where the structure breaks.
The Accountable Version
Here is what a falsifiable version of this claim would look like, and what you should demand before accepting any version of it.
State the year. State the price and the timestamp. State the base rate you are comparing against. State the drawdown threshold that would confirm the thesis and the one that would kill it. State the window: is the claim about the two weeks around November 3, or the full calendar year? Those are different bets and only one of them is interesting.
Then set the test. If 2026 produces a maximum intra-year drawdown below 27%, the decay trend is confirmed and the midterm-cycle thesis is falsified by its own author's data. If a drawdown does occur, check the date. If it lands in November, you have a coincidence worth investigating. If it lands in March, you have a macro liquidity event and a story that borrowed its credibility from a calendar.
Watch three things weekly, not annually: the drift in the 155-day cost basis as old coins age out, net ETF creation and redemption flows, and the options skew on the front-dated surface. Those are live. The election is a fixed constant that every participant has known about for two years.
Gravity doesn't negotiate with narratives. Neither does a base rate. The next midterm will produce a number, and the number will be quoted as confirmation no matter which direction it moves. The only real question is whether anyone reading it will still be asking for the denominator.