Why Do Crypto Prices Differ Between Exchanges and Trackers?

Introduction

Six major exchanges quoted bitcoin within $27.18 of each other at one instant. Two Korean exchanges, at the same instant, quoted it roughly $1,800 higher. Neither number is wrong, and the difference between those two facts is the whole subject. Crypto has no consolidated tape and no official price. There are only 2,375 separate BTC markets across 153 exchanges, each printing its own last trade, plus the trackers that compute one figure from them. This article measures how far apart real venues sit and shows what a plain average of every market would produce. It then opens up the filtering that prevents that result. The closing sections separate a price gap that signals a data fault from one that carries real information about liquidity.

Is There an Official Bitcoin Price Anywhere in the Market?

No authority publishes the price of bitcoin. There are only separate markets, each quoting its own last trade. Every number a reader encounters is either one of those quotes or a calculation performed over them. Equities have a consolidated tape. Crypto has 2,375 BTC markets and no referee.

No consolidated tape exists

United States equity markets run a consolidated tape: a single regulated feed that every venue reports into, producing one official last-sale price per symbol. Crypto has no equivalent and no regulator mandating one. CoinPaprika alone lists 2,375 BTC markets spread across 153 distinct exchanges, 2,357 of them spot and 18 derivatives (CoinPaprika API, 2026-09-10) . Each of those markets runs its own matching engine against its own order book ↗, and each prints a different last trade at any given moment. The absence is structural rather than temporary. Nothing in the design of a decentralised asset produces a single authoritative quote.

What a venue quote actually is

An exchange price is the last executed trade on that exchange, in that pair, at that moment. It carries no claim about the wider market. Two venues quoting different numbers are both reporting accurately about different order books. A tracker's number is something else entirely: a statistic computed across many venues, belonging to none of them. Confusing the two produces most of the confusion around crypto pricing. A reader compares an exchange quote against an index and reads the difference as an error, when it is two different measurements.

Data current as of September 2026.

The interesting question is not whether those six numbers differ, but by how little.

How Far Apart Do Major Exchanges Actually Sit Right Now?

Measured at a single instant, six major venues quoted bitcoin within $27.18 of each other. That is 0.035% of the price. The disorder most readers expect from crypto pricing does not appear on liquid venues. The reason it does not is worth more than the observation itself.

The live six-venue spread

At 13:18:06 UTC on 10 September 2026, Binance quoted $76,842.00 and Coinbase quoted $76,814.82. OKX, Bitstamp, Kraken and the CoinPaprika index all fell between them (CoinPaprika API, 2026-09-10; Binance API, 2026-09-10; Coinbase API, 2026-09-10) . The full range across all six was $27.18. For comparison, a single retail market order of moderate size moves the price on a thin venue by more than that. The spread between honest, liquid venues is smaller than the cost of trading on most of them. That is the practical definition of an efficient market.

Why the gap stays this small

Tight spreads across venues are not a courtesy. Market makers quote on several exchanges simultaneously and profit from any gap that opens between them, so the act of harvesting the gap closes it. The constraint is capital and speed rather than goodwill. When a venue's price drifts, the first firms with inventory on both sides pull it back within seconds. Bitwise made this argument to the SEC in 2019. Once fake volume is stripped out ↗, the real bitcoin market trades at a unified global price and is efficiently arbitraged (Bitwise Asset Management, 2019).

The number a tracker publishes, however, matches none of those six quotes exactly.

Why Does an Aggregator's Price Never Match Any Single Exchange?

A tracker's price is a computed statistic, not a quote from anywhere. It cannot match a specific exchange because it is not trying to. The naive alternative shows why the computation matters: a plain average across all 2,375 BTC markets returns $141,389.99 against a real price near $76,840.

An index is computed, not quoted

An index price answers a different question than an exchange price. The exchange answers "what did the last trade here cost". The index answers "what is this asset worth across the market as a whole". Those questions have different answers by construction, so expecting them to agree to the cent misunderstands both. The index also updates on its own schedule, blending venues that print at different rates. A reader comparing an index against one exchange at one moment is comparing a moving average of many books against a single book.

Mean versus median on 2,375 markets

The reason aggregators filter becomes obvious the moment anyone tries not to. Taking every one of the 2,375 BTC markets CoinPaprika lists and averaging their prices produces $141,389.99, roughly 84% above the real price. The raw distribution runs from $0.01 to $218,268.07, and a handful of broken or illiquid prints drag the mean upward on their own. The median of the same 2,375 markets is $76,879.26, within $41 of the published index of $76,838.36 (CoinPaprika API, 2026-09-10) . One statistic is unusable and the other is nearly right, computed over identical inputs. The contrast is not an argument against averages. It demonstrates that the input set, rather than the formula, decides whether a number is usable at all. A median survives bad inputs; an arithmetic mean does not.

Filtering is what separates the two, and the filters are visible.

How Do Trackers Turn Thousands of Markets Into One Number?

Aggregation is filtering first and weighting second. The weighting draws attention, but the filtering does the work of making the number usable at all. CoinPaprika's own market records expose that filtering directly, field by field, rather than describing it in a methodology page.

Filtering before weighting

A market has to earn inclusion before its price counts. Typical screens exclude venues with no verifiable depth and pairs whose quote currency has itself moved. They also drop prints far from the cross-venue consensus, and markets whose reported activity fails statistical plausibility tests. Only surviving markets reach the weighting stage, where volume determines influence. Running the weighting without the screening is what produces the $141,389.99 average. Volume weighting alone cannot rescue a dataset containing a $0.01 print. A venue claiming enormous volume would pull the result toward its own broken number.

What the API exposes about its own method

CoinPaprika's per-market records carry three fields that make the screening observable. Each market has an outlier boolean, a trust_score of high, medium, low or no_data, and an adjusted_volume_24h_share. Across the 2,375 BTC markets the trust scores split 883 high, 47 medium, 251 low and 1,194 no_data. Sixty-eight markets carry outlier: true, and every one of them is assigned an adjusted_volume_24h_share of zero (CoinPaprika API, 2026-09-10) . A reader can therefore inspect which markets were set aside rather than taking a methodology claim on trust. None of this makes a tracker's number authoritative. It makes the number auditable, which is a different and more useful property. Anyone who disagrees with a filter can recompute the figure from the same public records and see exactly where the two answers diverge.

Data current as of September 2026.

The screening explains why an index differs from any venue. It does not explain why some venues differ from each other by far more than $27.

What Keeps Prices Aligned Across Venues Most of the Time?

Arbitrage does not force prices to be equal. It forces them into a band whose width is set by fees, transfer time and price risk. The 0.035% spread measured across six venues is that band, and reading it as an absence of friction inverts the mechanism.

The arbitrage loop

The trade is mechanically simple. Buy on the cheap venue, move the coin to the rich venue, sell, and repeat until the gap closes. Each leg costs something: a taker fee on both sides, a network fee to move the asset, and a withdrawal fee at the exit. Those costs sum to a floor. No gap narrower than the floor is worth harvesting, so gaps persist at exactly that width and no narrower. The floor on liquid USD venues currently sits in the low tens of basis points, which is why the measured spread landed where it did.

What the bound actually costs

Time is the expensive input rather than fees. Moving bitcoin between venues takes blockchain confirmations, and during that window the price can move against the position. The arbitrageur cannot hedge the exposure by shorting on the rich venue, because the asset is not yet there to sell. Academic work on cross-border bitcoin premia finds exactly this pattern. Premia rise with transaction costs, median confirmation time and volatility, all three consistent with delay imposing unhedgeable price risk (Cong et al., 2019).

Remove the ability to move capital at all and the bound stops binding entirely.

When Do Arbitrage Bounds Break and Prices Stay Apart?

When capital cannot cross a border, the arbitrage loop cannot close. Korean venues quoted bitcoin 2.39% above the index at the same instant six USD venues sat 0.035% apart. The same mechanism once produced a gap above 50%, and eight years later it has still not gone away.

A live premium, measured today

At 13:18:06 UTC on 10 September 2026, Bithumb quoted 105,085,000 KRW and Upbit quoted 105,124,000 KRW. Converted at 1,336.20 KRW to the dollar, those are $78,644.66 and $78,673.85. Both sit above the CoinPaprika index of $76,838.36, by 2.35% and 2.39%, a gap of $1,806 and $1,835 per coin (Bithumb API, 2026-09-10; Upbit API, 2026-09-10) . The two Korean venues agree with each other to within 0.04%, which rules out a venue-specific glitch. They agree with each other and disagree with the world, which is the signature of a closed capital border rather than a data fault.

January 2018 and the limits of arbitrage

The same structure produced a far larger gap under stress. Between January 2016 and February 2018 bitcoin was on average 4.73% more expensive in Korea than in the United States. The gap peaked at 54.48% in January 2018 (Cong et al., 2019). Bloomberg reported the same episode as a 51% gap that closed to parity by 2 February 2018 (Bloomberg, 2018). Two credible sources differ by three percentage points because they used different venues and exchange-rate conventions. Even measurements of divergence are method-dependent. Bank of America later attributed the persistence directly to Korean capital controls that prevent effective arbitrage between onshore and offshore prices (Reuters, 2021). The premium has never been a constant. It flipped to a discount after Korea announced regulation in late 2018, and had fallen to roughly 2% by 2022 (AMRO, 2023).

Data current as of September 2026.

Borders are one reason two quotes disagree. The currency in the pair is another.

Why Do Stablecoin Pairs and Fiat Pairs Disagree?

A BTC/USDT quote prices two assets at once. Part of any gap against a BTC/USD quote is the stablecoin moving rather than bitcoin. A reader who assumes the quote currency is pinned to a dollar attributes that movement to the wrong asset.

The quote currency is a second asset

Tether is an asset with its own market, not a unit of account. When a BTC/USDT market prints 76,842, one bitcoin traded for 76,842 tether. Converting that into dollars requires knowing what tether was worth at that moment. Most of the time the conversion is close enough to ignore, since USDT trades near a dollar. The assumption becomes visible only when it breaks, and by then it has already been applied to every number derived from the pair. The convention hides the dependency well. Field names read price on both pairs, and nothing in a raw feed announces that one of them is quoted in an asset that can move.

When the stablecoin moves instead

During stress the quote currency moves first. A stablecoin trading at 0.99 makes every BTC pair denominated in it appear roughly 1% richer, even with bitcoin unchanged against the dollar. Aggregators convert each pair through the quote asset's own measured price rather than assuming parity. That is one more reason an index value differs from a raw venue quote. Comparing a USDT pair against a USD pair without that conversion measures the stablecoin as much as it measures bitcoin.

Those are structural reasons for divergence. The calculation itself can also move the market.

Can a Change in Index Methodology Move the Market?

A methodology change is not a neutral observation of the market. It changes the number everyone quotes. In January 2018 one such change moved that number sharply enough to move prices with it, and no trade was responsible for the drop.

The January 2018 exclusion

At the height of the Korean premium, CoinMarketCap removed Korean exchange prices from its global average. The change was not widely communicated, and the reported market price dropped abruptly as the elevated Korean quotes left the calculation (CoinDesk, 2018). CoinMarketCap was at the time the most cited price source in financial media, quoted by outlets from the Wall Street Journal to Barron's. Its average functioned as the market's reference number. Nothing had happened to bitcoin. The asset traded where it had traded a minute earlier on every venue. The most-quoted summary of its price fell, and participants reacted to the summary.

Why methodology is a market event

The episode establishes something readers rarely account for: aggregated prices are inputs to decisions, so changing how one is computed changes behaviour. Index construction determines which venues count, and that choice has consequences for liquidations, collateral valuations and automated strategies keyed to a published feed. Anyone depending on a single tracker inherits that tracker's methodology decisions, including future ones. The lesson generalises past that episode. Adding or removing a venue shifts the published price without a single trade occurring, and downstream systems cannot distinguish that shift from a real move.

The practical question is which number belongs in which place.

Which Price Should You Actually Use for Trading or Accounting?

Execution questions need the venue where the order will rest. Valuation, accounting and research need a filtered cross-venue index instead. Using either in the other's place is the real error, and it is both more common than a wrong number and harder to notice.

Execution price versus reference price

An order fills against one order book. For anyone about to trade, the only relevant price is the one on that venue, in that pair, at that size. No index will fill an order. For anyone valuing a holding, marking a portfolio or reporting a figure, the venue quote is the wrong choice. It reflects one book's momentary state and can be moved by a single participant. The distinction is between a price that can be transacted and a price that represents the market. Neither is more accurate than the other, because they answer different questions.

Matching source to job

The mismatch shows up in predictable places. Portfolio trackers using a single exchange quote misvalue holdings when that venue drifts. Backtests using an index ↗ assume fills at a price that existed on no venue. Tax and accounting work needs a documented, reproducible reference rather than whichever tab was open. Naming the source and its methodology alongside the number solves most of these problems at no cost. Recording when a figure was pulled helps as much as recording where it came from. That habit turns a later disagreement into a short reconciliation rather than an argument about whose number was right.

Data current as of September 2026.

Once the right source is chosen, a residual gap still needs interpreting.

How Do You Check Whether Two Price Feeds Disagree Legitimately?

Quote currency, venue set and timestamp explain almost every disagreement between two price sources. A gap that survives all three checks is usually reporting something real about liquidity ↗ rather than signalling a bug, and it deserves investigation rather than a support ticket.

Three checks before calling it a bug

The first check is the quote currency: a USDT pair and a USD pair are not the same measurement. The second is the venue set. An index across 153 exchanges and a single Korean venue answer different questions, and the difference between them is the answer. The third is the timestamp. Prices pulled seconds apart in a moving market differ for no interesting reason, and comparing a cached value against a live one manufactures a discrepancy.

When a gap is real information

A gap that survives those checks carries content. Persistent divergence on one venue points to thin depth, a withdrawal problem or a closed capital border. Each matters more than the price itself. The Korean premium is the clearest case. The 2.39% is neither noise nor error: it is the market pricing the cost of moving money across a border. Thin depth shows up the same way: a quote drifts and stays drifted, because no arbitrageur finds the size worth moving capital for. Reading either as a data fault discards the signal.

FAQ

References / Sources