In Miami, the hours before the open have a particular texture — the bay still holding yesterday's heat, the screens already cold and bright. On the last morning of September I sat watching two lines that the market had agreed to call a story.
The dollar index had risen roughly 2.6% since September 9, touching 101.69 — a two-month high. Bitcoin, over the same stretch, had slid from 87,500 into the 83,000–84,000 band, a drawdown of four to five percent. Two lines, opposite directions, and an industry that reached instantly for the oldest explanation it owns: the strong dollar is killing Bitcoin.
I have been watching this market for seventeen years, and I have learned to distrust explanations that arrive before the data does. In 2017, while finishing a master's in economics, I manually audited fifteen ICO whitepapers — reading them for the geometry of their tokenomics as much as for their code. The pattern never changed. The narrative preceded the evidence, and it was always cleaner than the evidence turned out to be.
So I went to the numbers. And the numbers, when you sit with them long enough, tell a quieter and more interesting story. A correlation is a memory of co-movement, not a mechanism of causation — and this particular memory is fainter than the market remembers.
Context: the dollar is a weather system, not a weight
The US Dollar Index is a trade-weighted basket — the euro carries the largest share, with the yen, the pound, the Canadian dollar, the Swedish krona and the Swiss franc filling out the rest. It is not a measure of American economic vitality. It is a measure of relative monetary conditions, and it rises when capital decides that dollar-denominated claims are the least uncomfortable place to sit.
The case for "strong dollar, weak Bitcoin" has a certain mechanical elegance. When the dollar appreciates, borrowers servicing dollar-denominated debt face higher real costs. Leverage contracts. Risk appetite thins. And because Bitcoin trades around the clock in a market with no circuit breakers and no central bank standing behind it, it becomes the first thing sold and the last thing bought. I have used a version of that chain in my own work. It is not wrong. It is just incomplete.
The specific data behind this debate came from TradingView — a single platform, which is worth noting. The dollar index, after its September run, had re-entered the Ichimoku cloud, the Japanese technical framework that treats price as a system of layered averages, and had not yet cleared resistance at 101.80. Bitcoin, for its part, had given back roughly a twentieth of its value without any corresponding collapse in on-chain activity that anyone bothered to cite.
My vantage point on all this is unusual, and I should be transparent about it. I work as a CBDC researcher at a Miami policy think tank, which means I spend my days in the narrow corridor between state-issued money and privately issued money — drafting frameworks, comparing prototypes, arguing about where the boundary belongs. The dollar is not an abstraction to me. It is a design surface. And when I see a single exchange-rate index being used to explain the price of a globally distributed bearer asset, I want to know what is being measured and what is being assumed.
Core: the coefficient is a headline, the R-squared is the finding
Here is what the numbers actually say.
Over the past ninety trading days, the correlation between the dollar index and Bitcoin's daily returns was -0.41. That figure has been widely circulated, because it is the lowest reading since February 2023 — a fact that flatters the bearish narrative. Negative correlation, the story goes, and a multi-year extreme.
But a correlation coefficient is not a result. It is a ratio, and the ratio only becomes a result when you square it. The coefficient of determination here is 0.17. In plain terms: the dollar index explains roughly seventeen percent of Bitcoin's daily variance. The other eighty-three percent belongs to something else.
That distinction matters more than any single number in the dataset. A -0.41 correlation sounds like a relationship. An R-squared of 0.17 sounds like an acquaintance. They are the same statistic, reported by two different temperaments, and only one of them is honest about what the market can actually trade on.
Two days that carried a month
The thirty-day window is where the analysis earns its keep. On a rolling thirty-day basis, the correlation reads -0.45 — tighter, more convincing, more quotable. Then you remove two sessions: August 19 and September 3. Both were outlier days, the kind of prints that occur when a single large liquidation cascade or a macro headline lands on thin liquidity.
Strip those two days out, and the thirty-day correlation falls to -0.19, with an R-squared near 0.04. Four percent. That is not a relationship between two assets. That is two objects briefly sharing a room.
A coefficient that swings from -0.45 to -0.19 on the removal of two sessions is not a parameter — it is a weathervane. It spins with the wind and it tells you which way the wind blew, never which way it will blow. Anyone building a position on the strength of a thirty-day reading is building on the two most unusual days of that month.
This is exactly the kind of test I would run on a protocol's claimed performance metrics before writing a word about it. Take the headline number. Find the observations that carry it. Remove them. See what remains. Most of the time, what remains is much smaller, and much truer.
The long view refuses to cooperate
Step back further and the case weakens again. Since January 2020, the ninety-day rolling average correlation between the dollar index and Bitcoin sits at -0.14. Weak. Persistent in sign, negligible in magnitude. If the dollar were the gravitational force in Bitcoin's pricing, five years of daily data would have surfaced it by now. Gravity does not take semesters off.
And then there is November 2024, when the same ninety-day measure turned positive, at +0.22. Dollar up, Bitcoin up — the two lines moving together. That single episode is the most instructive data point in the entire set, because it destroys the assumption that the relationship has a fixed sign. It does not. It has a regime.
The dollar is not a weight on Bitcoin; it is a weather system — and weather does not push, it conditions. In November 2024, the market was pricing a coherent macro thesis in which dollar strength and risk-asset strength were the same trade. In September 2025, it was pricing something closer to a defensive rotation. Same two variables. Opposite sign. A correlation coefficient cannot hold that, and pretending otherwise is how models get blown up.
What the analysis did not check
The report's other finding deserves more attention than it received: Bitcoin showed no significant correlation with US Treasury yields either. Read that alongside the dollar result and a picture forms. The two most-cited macro anchors for risk assets — the dollar and the risk-free rate — both fail to explain Bitcoin's daily behaviour in any durable way.
What does explain it? The honest answer is a composite. Global liquidity conditions, which move slower than the dollar index and are harder to quote on a chart. Crypto-native flow, which in the post-ETF era means the daily creation and redemption activity of a handful of large vehicles. The supply schedule, which is fixed and known and therefore never a surprise. And a general risk-appetite factor that the dollar index partially, imperfectly proxies.
But I want to push past what the report measured and name what it skipped, because this is where my own work has lived for the past three years.
The first omission is driver heterogeneity. The report treats "a strong dollar" as a single, undifferentiated condition. It is not. A dollar rising because American growth is outrunning the rest of the world is a risk-on dollar, and Bitcoin has historically tolerated it. A dollar rising because offshore funding is seizing up is a risk-off dollar, and that one does real damage. Identical prints on the index, opposite implications for a bearer asset with no domestic economy behind it. The reason for the move is a more important variable than the move itself.
The second omission is the plumbing — the channels through which the dollar actually touches this market, none of which appear in a daily-return correlation matrix. Bitcoin miners borrow in dollars to buy machines; a stronger dollar raises their real servicing cost and can force coin sales that look, from the outside, like sentiment. Stablecoin issuance and redemption are dollar-denominated operations; when the dollar tightens, the mint-and-burn mechanics that lubricate crypto liquidity slow down. And in the institutional era, Bitcoin increasingly sits inside multi-asset portfolios governed by volatility-targeting or risk-parity mandates, which mechanically reduce exposure when realised volatility rises — a transmission path that is nonlinear, slow, and completely invisible to a ninety-day coefficient.
The third omission is the structural break hiding in plain sight. The report notes that -0.41 is the lowest reading since February 2023. That is not a footnote; it is a timeline. Between February 2023 and September 2025, spot Bitcoin ETFs launched, institutional custody matured, and the derivatives complex deepened enormously. The composition of who owns Bitcoin changed. If the dollar's grip on Bitcoin has loosened, that is the most plausible reason — and it is a structural story, not a statistical one. A correlation study can detect the change. It cannot explain it.
Contrarian: weak correlation is not independence, and it is certainly not safety
I think the report's conclusion is probably right — the dollar's suppression of Bitcoin is overstated — and I think it is right for a fragile reason, which makes it dangerous.
Correlation is a scalar. It compresses every state of the world into a single number between -1 and +1. But relationships between assets are not scalars; they are conditional. They behave one way in calm markets and another way in a funding squeeze. They behave one way when the dollar rises on growth and another when it rises on fear. A coefficient estimated across a mixed sample will report the average of two different worlds and describe neither.
The tail is where this matters. In March 2020, correlations across every risk asset converged toward one — the moment when diversification stopped being a strategy and became a liability. I spent much of 2022 drafting a fifty-page internal memo on exactly this dynamic, mapping how macro liquidity cycles dictate the specific shape of crypto collapses: the slow compression first, then the cascade, then the long grey silence while leverage unwinds. Nothing in that memo relied on the dollar index. Everything in it relied on the fact that when liquidity leaves, it leaves everywhere at once.
So here is the contrarian read. The risk is not that the dollar is a constant weight on Bitcoin. The risk is that it becomes one abruptly, at the exact moment its reason for rising changes. A report concluding that the threat is overestimated is arithmetically sound and strategically incomplete. It measures the average case and says nothing about the case that matters.
There is a second thing the analysis leaves out, and it is the thing I would have flagged first. Bitcoin has an internal supply mechanism — a fixed issuance schedule, a known halving cadence, and an on-chain holder base whose behaviour is observable in real time. None of that entered the framework. The report treats Bitcoin as a price series attached to a macro variable, which is a legitimate analytical choice and also a choice that quietly assumes the asset has no interior. A price is a rumour that liquidity agrees to believe, briefly — but underneath the rumour, there is a ledger, and the ledger has a structure. Ignoring it means the analysis can tell you what Bitcoin correlates with and never why it moves.
And one more caution, aimed at my own side of the aisle. In a bull market, weak-correlation findings get repurposed as permission. The story becomes: the dollar cannot hurt us, so the position can stay on. That is not what the data says. The data says the dollar has not been the dominant driver. Those are different claims, and the gap between them is where portfolios die.
Takeaway: watch the threshold, and watch the coefficient change character
There is a specific number to watch, and it is not a correlation coefficient. It is 101.80 — the resistance level the dollar index has not yet cleared. Not because round numbers are magic, but because that is where the consensus will either confirm itself or break. A clean break above it forces a re-pricing of the dollar leg across every risk book on the planet, and the question of whether Bitcoin is correlated will stop being academic.
There is also a subtler signal, and it is the one I will be tracking personally: not the level of the Bitcoin-dollar correlation, but its character. A coefficient drifting toward -0.6 or beyond would tell me the macro factor is gaining authority over this market — that Bitcoin is being traded as a risk asset again, in the plainest sense. A coefficient oscillating near zero tells me something more interesting: that this asset is developing its own interior, its own supply-demand logic, its own weather.
I do not think Bitcoin has decoupled from the dollar. I think it is differently coupled — sometimes through the front door of daily returns, more often through the back channels of miner debt, stablecoin mechanics and institutional rebalancing that no ninety-day window will ever show you. A transaction is just a promise frozen in time, and promises are priced by whoever is doing the promising — not by the currency they happen to be denominated in.
So the question I am left holding, on the last day of September, is not whether the strong dollar threatens Bitcoin. It is this: if the dollar's next leg up comes not from American growth but from a global scramble for funding, will we still be able to call the threat overestimated — or will we discover that we were measuring the wrong thing all along, precisely because the right thing was quiet?
