Why chart pattern matching should stay on one exchange
The same Bitcoin chart search scored 50.67% on Binance and 47.33% on Coinbase. Why Comparand matches a coin only with its own past on one exchange.
Bitcoin on Coinbase and Bitcoin on Binance are not the same price series, and the same chart search gave each its own result. In 300 searches each, the ten closest matches called the next-day direction right 50.67% of the time on Binance BTCUSDT and 47.33% on Coinbase BTC-USD, a gap that chance alone can make at 300 tests. Underneath, Coinbase BTC-USD spot was a median 3.26 bp above Binance BTCUSDT perps. A bp (basis point) is one hundredth of one percent, so that is about $3.26 per $10,000.
That is why Comparand compares a coin only with its own past on the same exchange. These are past cases, not a forecast.
For why the two prices differ at all, see Why is Bitcoin price different on exchanges?.
What does a chart search do?
Comparand's chart-shape search takes the last stretch of completed bars for one market. A bar records the open, high, low and close price over a set time. The search turns that stretch into a shape, then finds the closest past stretches of the same coin on the same exchange and timeframe.
For each past match, it shows the price path and what the price did next, in bp, at one quarter, one half and the full length of the window. It does not average the results into a score.
A "market" here means the exchange, the coin and the type of contract. BTC on Binance is one market. BTC on Coinbase is another. Spot means trading the coin itself. A perp is a futures contract that never expires.
How far apart are the prices on two exchanges?
We paired hourly closes with the same UTC time label and computed the gap for each. UTC is the world time standard. A positive gap means Coinbase's number was higher.
| Period | Paired hourly closes | Median gap | 5th percentile | 95th percentile |
|---|---|---|---|---|
| Whole period, Sep 2019 to Oct 2026 | 62,342 | +3.26 bp | −12.36 bp | +15.05 bp |
| 2020 | 8,781 | +2.83 bp | −22.82 bp | +25.14 bp |
| 2025 | 8,755 | +5.33 bp | −3.95 bp | +11.13 bp |
| 2026, to Oct 20 | 7,004 | −1.01 bp | −11.07 bp | +8.29 bp |
A percentile is a cutoff in the sorted gaps. About 5% of gaps fell below the 5th-percentile number and 5% above the 95th. They are not the smallest and largest gaps.
The middle gap is small. The tails are wider, and the gap changes direction. In 2026 the median went negative. In 2020 the tails stretched from −22.82 to +25.14 bp.
Our companion article, why Bitcoin's price differs between exchanges, explains where the gap comes from. In short, it mixes the exchange, the currency (USD or USDT) and the contract type (spot or perp).
Why does a price gap matter for a chart shape?
A shape is built from price changes, not price levels. A steady gap, such as a fixed 3 bp, would not change a shape.
But the gap is not steady. It sits near the median in some hours and far from it in others. We measured how wide it ranged, not how much it changed hour to hour, so we cannot say how often it bends a shape. We can say it is not zero, and it differs from year to year.
A shape search also compares many pieces at once. Small differences in the pieces can change which past stretches look closest.
Do two exchanges give the same search results?
We ran our own test of the chart search on Bitcoin and Ethereum, on each exchange separately. For each market we ran 300 searches at 300 evenly spaced past times, and compared the median of the 10 closest matches with what price did 24 hours later. We never mixed markets.
| Market | Next-day direction matched | "Always up" at the same times | Searches where the 10 matches split 4 to 6 up |
|---|---|---|---|
| Binance BTCUSDT perps | 50.67% | 49.33% | 65.00% |
| Coinbase BTC-USD spot | 47.33% | 49.00% | 64.33% |
| Binance ETHUSDT perps | 51.33% | 54.33% | 64.33% |
| Coinbase ETH-USD spot | 50.33% | 47.33% | 63.33% |
"Always up" is the score you would get by guessing a rise every time, at the same 300 times. The search beat it on two markets and lost on two. A "4 to 6 up" split means the 10 past cases disagreed about direction. All four markets were near a coin flip, and across 12 markets the matches picked the next-day direction 50.22% of the time, below the 50.92% from always guessing up. Our article on whether chart look-alikes predict crypto has the full result.
The scores differ by a few points, but at 300 tests each, that is within what chance can do. So we do not say Coinbase and Binance behave differently in the search. We only say that each market gave its own result, and nothing in the test lets us treat them as one sample.
How much do the closest matches move even on one exchange?
We also changed the chart length from 48 to 96 hourly bars and counted how many of the 10 closest matches stayed the same. The same match means the same past ending time.
| Market | Matches that stayed (of 10) | Searches where none stayed |
|---|---|---|
| Binance BTCUSDT perps | 1.96 | 35.00% |
| Binance ETHUSDT perps | 1.90 | 29.67% |
| Coinbase BTC-USD spot | 1.88 | 39.00% |
| Coinbase ETH-USD spot | 1.52 | 42.33% |
On a single exchange, changing one setting kept only about 2 of the 10 closest matches. Our guide on how chart length changes the matches goes further.
We did not test a mixed pool of exchanges, because Comparand does not build one. But a match list is clearly sensitive. Adding another exchange's different prices would be one more change that the reader could not see inside the list.
Do exchanges have the same amount of history?
No. The history length and the missing hours differ, and a search can only look back as far as its exchange goes.
- Coinbase hourly data starts in January 2017. Binance perps start in September 2019, OKX in January 2020, Bybit in March 2020, and Hyperliquid in April 2025.
- In the hourly price records we tested, Coinbase BTC-USD had 55 missing hours and Coinbase ETH-USD had 213. Binance BTCUSDT and ETHUSDT had none.
A missing hour is a gap in the record. We never fill it with an invented price. In our tests, any window or outcome that touched a missing hour was left out.
So the choice of exchange also sets how many past cases exist for you to see. Coinbase's history starts earlier, so a Coinbase search can find cases from 2017 to 2019 that a Binance search cannot.
How should you pick the exchange for a search?
Use the exchange whose chart you are looking at, or the one where you would actually trade. Then check the full market name: exchange, coin and type. BTC-USD spot and BTCUSDT perps are different markets. Our guide to perpetual futures explains how a perp differs from spot.
To see another exchange, run a separate search there and read it as that exchange's own past cases.
How we measured this
We paired completed hourly closes from Coinbase BTC-USD spot and Binance BTCUSDT perps with the same UTC close time, from September 2019 to October 20, 2026, and divided the Coinbase close by the Binance close. We did not convert USDT to dollars, and the gap leaves out fees, funding and transfer limits. For the matching test, we used 48 hourly bars, 10 matches and 300 searches per market. For the length test, we used the same 300 times for each market. Searches that overlap in time share price history, so the 300 tests are not 300 independent trials. We tested only Bitcoin and Ethereum for the exchange comparison.
Try this in Comparand
Ask your connected AI app: "Which exchanges do you cover, and which Bitcoin markets does Coinbase list? Then find the 10 closest past charts to the latest 48 hourly bars on the exchange I name."
- venue_coverage lists the other exchanges and prediction markets served, with the data and time steps each one has. It is free.
- venue_markets lists the markets one exchange has, so you can pick the exact symbol and type. It is free.
- shape_analogs takes the exchange (venue) and symbol, plus an optional timeframe, window length (20 to 200 bars, default 48) and number of matches (1 to 20, default 10). It returns each match's path and forward move. It never returns matches from another coin or exchange, gives no average or hit rate, and costs $0.01 per search.
For Hyperliquid perps, set the venue to hyperliquid. You can try 3 coins free, read the data tool docs, or see pricing. To learn what an analog is, start with what a chart analog means.
The short version
- Coinbase BTC-USD was a median 3.26 bp above Binance BTCUSDT over 62,342 hours, with a 5th-to-95th range of −12.36 to +15.05 bp.
- Exchanges also differ in currency, contract type, history length and missing hours.
- Our 24-hour tests gave different scores per market, but 300 tests each is too few to rank them.
- Changing chart length kept only about 2 of the 10 closest matches, even on one exchange.
- Comparand compares a coin only with its own past on the same exchange. Past cases, not a forecast.
Questions people ask
Why does Comparand compare a coin only with its own past on the same exchange?
Each exchange has its own prices, products and gaps in its history. Coinbase BTC-USD spot was a median 3.26 bp above Binance BTCUSDT perps over 62,342 hourly closes, and the gap ranged from -12.36 to +15.05 bp at the 5th and 95th percentiles. Keeping a search on one exchange avoids mixing those differences into a chart shape.
How big is a basis point?
A basis point (bp) is one hundredth of one percent. So 3.26 bp is about 0.03%, or about $3.26 on every $10,000.
Do Binance and Coinbase give the same chart search results for Bitcoin?
No. In our 24-hour test with 300 searches per market, Binance BTCUSDT matched the next day's direction 50.67% of the time and Coinbase BTC-USD 47.33%. A gap that small is within what chance can produce at 300 tests, so we do not claim one exchange is better.
Can I run a chart search on a different exchange?
Yes. The chart-shape search takes a venue and a symbol, and it works on Binance, Bybit, OKX and Coinbase for big coins, plus Hyperliquid perps. Each search stays on the exchange you choose.