Coinbase Premium Index

The Coinbase Premium Index plots the daily percent spread between Bitcoin's US-regulated Coinbase USD price and its largest offshore USDT price, snapshotted once per UTC day. Rust fill above the zero baseline marks days when Coinbase printed higher — US bid dominant. Slate fill below marks days when offshore printed higher — US flow skewed seller. A muted Bitcoin price line sits behind on the left log scale for cycle context.

Today's raw daily reads −0.054%. The seven-day rolling mean is −0.068% — Balanced. The thirty-day mean sits at −0.100%, across 3,000 daily observations going back to February 2018. What makes the spread legible is the asymmetry of its two legs: one venue clears in US dollars under US regulation, the other in a dollar-pegged stablecoin offshore. The gap between them is, mechanically, a price on where the dollar wants to enter Bitcoin.

Take the day's Coinbase BTC-USD close and the same day's offshore BTC-USDT close, compute the spread as a percent of the offshore price, and quote the result. The construction follows the canonical platform definition; our pipeline carries the percent series end-to-end and exposes the dollar counterpart (sometimes called the Coinbase Premium Gap) in the tooltip. The formula:

premium_rate = (coinbase_USD − offshore_USDT) / offshore_USDT × 100

A unit-convention note, because it is the single most common misreading. The upstream premium_rate field is already quoted in percent, not as a fraction — despite what the English word “rate” suggests. A value of 0.05 means Coinbase printed 0.05% higher than offshore, not five percent. The convention is preserved end-to-end so the regime thresholds (±0.10% / ±0.25%) are read literally. Full provenance lives on the data sources page and the per-step reconstruction in the methodology.

The seven-day mean carries the regime-classification work because Coinbase's intraday outages print as instantaneous spikes in either direction with no flow behind them. The 19 May 2021 sell-off — the China-crackdown capitulation day — left both Coinbase and the largest offshore venue partially down for hours; the official incident archive is the canonical record of the rest. The seven-day average smooths these one-day discontinuities into background; the raw daily print stays on the chart for texture.

The seven-day mean sorts into five bands. Above +0.25%, US-side buying is heavy — multi-day clusters of this size historically bracketed accumulation phases ahead of major price advances. Above +0.10%, US institutional accumulation is the working phase. Inside ±0.10% is the balanced band, where neither side of the cross-venue book is dominating, and where most of the post-2024 record lives — not because demand vanished, but because the demand migrated off the lit book the index reads. Below −0.10%, US flow has flipped seller; below −0.25% is the smart-money-offloading register that has flagged both cycle distribution windows and routine institutional rebalancing.

Seven anchor dates from across the indicator's life make the regime rotation visible. Each row pins a named market event; the premium-rate cell and the spot price are pulled from the daily snapshot powering the chart above — so the table is a live read of the indicator's record, not a frozen retelling. The sharpest pre-2024 anchor is the FTX-week panic, which prints deep into the negative offloading band; the January 2021 rally reads only mildly positive, and the post-ETF anchors print quiet by comparison — which is the point of the section below.

The premium has not measured the same thing across every cycle. Looked at by information content rather than live print — how much each reading is worth, not what it says — three eras separate cleanly.

2018–2020 — the pre-IPO US-onramp era. Coinbase was the only regulated path to a Bitcoin position for most US institutional capital. A persistently positive premium meant US balance sheets were accumulating on the only book they could legally clear. The contemporaneous read from January 2021 framed the regime cleanly: the premium running above the equivalent of +$50 bracketed every break above $20k, $30k and $40k in the months that followed.

2020–2024 — the duopoly era. Coinbase IPO'd on 14 April 2021 and the offshore venue's US arm was a meaningful retail onramp until the November 2023 enforcement settlement forced its wind-down. During that window the cross-venue spread captured a US-vs-rest signal cleanly, but its absolute level was no longer apples-to-apples with the pre-IPO record. Some of the US bid was clearing inside the offshore venue's US affiliate and never showed up in the spread.

2024–present — the post-spot-ETF era. US institutional flow now creates ETF shares at NAV. The authorised participants source the underlying through Coinbase Prime OTC desks, which clear at the daily benchmark fixing rather than the lit Coinbase orderbook the premium measures. The signal still works — an ETF issuance day with strong inflows still drags the premium positive — but at lower amplitude, and with the understanding that the institutional flow doing the work is the part that doesn't route through APs. Independent market-structure analysis documents the corresponding shift in benchmark-fixing concentration and the rise of US-exchange depth share to roughly 45% post-launch.

The Coinbase Premium is one of the most structurally fragile indicators on this site — not because the arithmetic is wrong, but because the two orderbooks it subtracts have been quietly redefined three times. Each break below has a date and a mechanism.

The NAV bypass (11 January 2024 onward). US institutional accumulation now routes through ETF authorised participants who create shares at NAV via Coinbase Prime OTC desks — flow that does not touch the lit Coinbase orderbook the premium measures. The December 2024 analyst quicktake flagged the resulting pattern: the premium closed the year sub-zero, with the 31 December 2024 print sitting in the institutional-selling band — while the spot ETFs were absorbing tens of thousands of coins that quarter. The premium read distribution; the custody data read accumulation. Both were true of different books.

The venue-mix redefinition (November 2023). The offshore venue's geographic mix changed materially after the November 2023 settlement forced the wind-down of its US affiliate. Pre-2023, some US flow cleared inside that US affiliate and muted the apparent US-vs-rest gap; post-2023 the gap reads cleaner but its absolute magnitude is on a different basis. A +0.10% print in 2021 and a +0.10% print in 2025 are not the same fact. Compare regime, never level, across that boundary.

The single-day outage artefact (recurring). Coinbase has had documented intraday outages; the 19 May 2021 sell-off is the textbook case, with both venues running degraded as Bitcoin fell roughly thirty percent intraday. The raw daily premium on those days carries snapshot artefacts from one venue's partial book, not flow. The seven-day mean exists to absorb them; never act on a single day's extreme. Taken together, these three breaks mean the honest read is directional and regime-scoped — the index tells you which way the lit US book is leaning this fortnight, and nothing about the parts of US demand that have moved off it.

If you are timing the cycle, treat sustained extremes contrarian-skewed in the post-spot-ETF era. A multi-week mean above +0.10% still has the same directional flavour it did pre-2024, but the path-dependent return has compressed because the fastest US bid is no longer in this print. Pair it with ETF flows: if both lines are positive you have one signal; if they disagree, the story is on the ETF side, and the premium is reading the wrong book.

If you are working the data, the percent series, the seven- and thirty-day means, and the dollar-gap counterpart in the tooltip are the inputs. The seven-day mean is the regime classifier specifically because raw daily readings catch venue outages; the thirty-day baseline lets you separate fortnight-scale moves from cycle-scale ones. Cross-reference exchange reserves to see the custody redirection the spread now misses, and treat any reading across the November 2023 boundary as belonging to a different basis.