What Is OHLCV? The Data Behind Every Crypto Chart

Introduction

Bitcoin moved $1,888.70 on 9 September 2026 and closed $174.92 from where it opened — a range 10.8 times the candle body. Anyone reading the open and close alone would record a flat day. That gap sits underneath every chart, indicator and backtest in crypto. OHLCV — open, high, low, close and volume — is a compression format. Compression discards the order trades arrived in, the spread between buyers and sellers, and the depth behind every quote. This article covers what each of the five values records and how exchanges build a bar from raw trades. It also measures why CoinPaprika, Binance and Kraken publish three different candles for the same day. The closing sections list the data-quality failures that corrupt datasets before analysis begins.

What Is OHLCV Data and What Does Each of the Five Values Mean?

OHLCV is one record of five numbers plus a timestamp: open, high, low, close and volume. Every candlestick chart, every indicator and every backtest in crypto reduces to that structure. The format stays simple by discarding almost everything that happened inside the period it describes.

The five values, field by field

Those five fields plus a timestamp are the whole of OHLCV data. The open is the first trade price inside the interval and the close is the last. High and low mark the extreme trade prices reached at any point between them. Volume totals the quantity that changed hands, and the timestamp identifies which interval the record covers. Exchanges publish these as arrays rather than objects. Position carries meaning, and a misread index shifts a price into a volume column without raising an error. The close does most of the analytical work downstream. Moving averages, the relative strength index (RSI) and Bollinger Bands all read the close by default ↗. One of the five OHLCV values ends up disproportionately responsible for what indicators report.

One real bitcoin candle, measured

CoinPaprika's daily OHLCV record for bitcoin on 9 September 2026 opened at $78,441.43 and closed at $78,266.50. In between it reached $79,697.25 and fell to $77,808.54, on $30.0B of volume (CoinPaprika API, 2026-09-10) . The distance from high to low was $1,888.70, or 2.41% of the open. The distance from open to close — the candle body — was $174.92, or 0.22%. The range ran 10.8 times the body. Bitcoin travelled roughly $1,890 across the day and finished $175 from where it started. A reader checking open and close alone would record a flat session. The wicks hold the rest.

Data current as of September 2026.

OHLCV values arrive already processed, and the processing rule determines what they say.

How Do Exchanges Build a Single Candle Out of Raw Trades?

An OHLCV bar is constructed rather than measured. An exchange collects every trade whose timestamp falls inside a window. The first sets open and the last sets close, the extremes set high and low, and the sizes sum into volume. The construction rule decides what the candle reports.

From trade tape to one bar

The raw input is the trade tape: one row per executed transaction, each carrying a price, a size and a timestamp to the millisecond. A liquid pair prints thousands of these per second. Binance recorded 3,338,369 individual trades in bitcoin against tether on 9 September 2026. Aggregation compresses all of them into one OHLCV row, and the compression ratio is the entire appeal. A year of daily bars runs to 365 rows. A year of trades on a major pair runs to billions. Nothing in the operation requires judgement. Four of the five OHLCV values are order statistics on the same list of prices, and the fifth is a sum.

Where the interval starts and stops

Bar boundaries align to clock multiples. An hourly bar covers 14:00:00.000 through 14:59:59.999, and a five-minute bar starts at :00, :05, :10 and so on. Intervals are half-open. A trade stamped exactly 15:00:00.000 belongs to the next bar rather than the one that closed (SiftingIO, 2026). Providers that rebuild OHLCV from raw trades emit nothing for an interval with no trades instead of carrying the previous close forward. Kraken states this directly: a missing candlestick in its downloadable history means no trades occurred, not that data went missing.

The boundary that closes one bar and opens the next turns out to be the most common source of broken OHLCV datasets.

Why Do Candle Timestamps and Bar Boundaries Break So Many Datasets?

Two OHLCV feeds can agree on every trade and still disagree on every bar. They differ on whether a bar's timestamp labels the start or the end of its window. A join assuming the wrong convention shifts every downstream indicator by exactly one bar.

Start-stamp or end-stamp

Binance identifies each kline by its open time and returns the close time as a separate field. A daily bar ends at 23:59:59.999 (Binance Spot API docs, 2026). CoinPaprika returns time_open and time_close explicitly, closing its 9 September daily candle at 23:59:59Z. Kraken returns a single timestamp per row and leaves the convention to the reader. Three OHLCV formats, three conventions, one asset. The mismatch survives every sanity check that plots the series, because a chart shifted by one bar still looks like a chart. It fails in a join, a merge or a backtest.

Why an off-by-one join shifts everything

An indicator reading a close that belongs to the following period is reading the future. A moving average built on misaligned bars produces signals nobody was able to act on. The error compounds quietly: backtest returns improve, which reads as a better strategy rather than a broken join. Daily bars carry a second trap. A daily bar for a 24/7 crypto market is a midnight-to-midnight slice of whichever timezone the provider chose (SiftingIO, 2026). Two providers cutting on different clocks produce different opens for the same calendar day.

Timestamps explain part of the divergence between OHLCV feeds. The volume column explains a larger part.

What Does the Volume Number in an OHLCV Record Actually Count?

The V is the least standardised letter in OHLCV. Binance reported 14,129.92 BTC for 9 September 2026 while CoinPaprika reported $30.0B for the same day (Binance Spot API, 2026-09-10; CoinPaprika API, 2026-09-10). Both figures are correct. They count different units across different venue sets.

Base volume versus quote volume

Base volume counts the asset itself: 14,129.92 bitcoin changed hands on Binance's BTC/USDT pair that day. Quote volume counts the currency paid for it, $1,114,730,666 on the same pair and the same day (Binance Spot API, 2026-09-10). Dividing one by the other returns the volume-weighted average price of $78,891.50, a number that appears in neither the open nor the close. Kraken publishes that average as its own column and reported $78,698.40 for the identical day. CoinPaprika reports volume in dollars only. Three feeds, three units, and no OHLCV field name that reliably separates them.

Why one venue is not the market

Scope matters more than units. Binance's 14,129.92 BTC covers one pair on one exchange. Kraken's 2,266.06 BTC covers one pair on another. CoinPaprika's $30.0B aggregates every venue it tracks. That scope is why it ran 26.9 times Binance's quote volume for the identical day (as of 9 September 2026) . None of those figures is a market total in the sense a stock exchange reports one. Crypto has no consolidated tape, so every OHLCV volume figure carries an implicit venue list that the field name never states.

Units and scope explain what volume measures. The deeper limitation sits in what the other four OHLCV values leave out.

What Can a Candle Never Tell You About What Happened?

Aggregation is lossy by design. OHLCV preserves four prices and a sum. It discards the order in which trades arrived, the spread standing between buyers and sellers, and the depth behind each quote. Two opposite markets produce a byte-identical record.

Two opposite paths, one identical record

A five-minute bar that opens at $100, prints $105, falls to $98 and closes at $101 describes two entirely different markets. In the first, price spiked immediately and bled out over four minutes. In the second, it sold off first and recovered in the closing seconds. Both produce the same open, high, low and close. An indicator reading that bar cannot separate them, because OHLCV discards trade ordering at construction (CoinAPI, 2025). Momentum, mean reversion and breakout logic all consume the same ambiguous input. The candle records where price went, never the sequence it took. The ambiguity is not an edge case. Any bar holding more than one trade admits multiple orderings, and every crypto bar above one second does.

What the candle never carried

Bid-ask spread never enters the record. Neither does order book depth, so a bar showing $30.0B of volume reveals nothing about the size fillable at any given price ↗. Cancelled orders leave no trace. Nor does the split between aggressive buyers and sellers, unless a provider ships it as an extra field. Binance does, as taker buy base and taker buy quote volume — fields standard OHLCV omits. Volume states a total and never a distribution. The same figure covers three institutional block trades and 50,000 retail transactions. Backtests built on OHLCV ↗ inherit three assumptions the data cannot support: perfect fills at the close, unlimited liquidity at the low, and zero spread throughout.

OHLCV compression removes detail at a fixed rate, and the interval setting controls how much.

How Does Changing the Timeframe Change What the Same Data Shows?

Timeframe is the compression setting on OHLCV data. Minute bars rebuild into hourly bars without loss, and hourly bars never rebuild into minutes. Each step up the interval ladder destroys detail no downstream indicator can recover. Interval choice is a data decision before it becomes a strategy decision.

Resampling up is safe, down is impossible

Aggregating minute bars into an hourly bar follows the rule that built them: first open, last close, maximum high, minimum low, summed volume. The operation is exact. The reverse has no solution, because an hourly bar holds four prices and no arithmetic recovers the 60 minute-bars behind them. Storage carries the same asymmetry — 30 days of bitcoin at one-minute resolution is 43,200 OHLCV rows against 30 rows at daily resolution. Teams that store only the coarse series meet the constraint when a strategy needs finer granularity. Partial periods compound it, because the newest hourly bar closes on incomplete input.

Choosing an interval without fooling yourself

Short intervals carry more detail and more noise. A one-minute candle registers micro-movements that say nothing about trend, while a daily candle smooths intraday moves a stop-loss would have hit. Neither is more accurate. Most strategy development starts at hourly resolution as a compromise between granularity and noise (as of September 2026) . The interval also sets what a pattern means: an engulfing candle on a weekly chart and on a five-minute chart describe unrelated events. Volume behaves the same way, and a single spike spread across 60 one-minute bars disappears into one hourly total.

Data current as of September 2026.

Interval choice changes the picture within one OHLCV feed. Switching feeds changes it again.

Why Do Two Providers Report Different Candles for the Same Day?

Three causes explain nearly all disagreement between OHLCV providers ↗: venue coverage, aggregation method and clock alignment. The gap is measurable rather than theoretical. The same bitcoin day differed by $62.75 at the high and $39.93 at the close across CoinPaprika, Binance and Kraken (CoinPaprika API, 2026-09-10; Binance Spot API, 2026-09-10; Kraken REST API, 2026-09-10).

Coverage, method, clock

No single feed sees every trade. A provider ingesting eight venues and one ingesting three disagree on volume by construction. If the high of a minute printed on a venue only one of them covers, they disagree on the high as well (SiftingIO, 2026). Aggregation method separates them further. A single-venue passthrough reports one exchange's tape including stale prints, while a cross-venue index reconciles disagreeing sources through a weighted median with outlier scoring. Clock alignment finishes the job: trades near a boundary land in different bars when one provider cuts on UTC and another on venue-local time.

The 9 September test across three sources

Pulling the same UTC day of OHLCV from three independent endpoints quantifies the effect . Opens spanned $14.37, highs $62.75, lows $53.64 and closes $39.93 — each under 0.08% of price, and each wide enough to move a stop-loss trigger. A second difference matters more. On both single-venue feeds the close chained exactly into the next day's open: $78,306.43 on Binance, $78,288.60 on Kraken. CoinPaprika's aggregated series opened $19.42 away from its own prior close. An index is recomputed across venues rather than carried forward from a last trade.

Data current as of September 2026.

Disagreement between honest OHLCV feeds is one problem. Data that is wrong inside a single feed is another.

Which Data-Quality Failures Should You Check For Before Trusting Candles?

Six failure modes account for most corrupted crypto OHLCV datasets. The Binance.US print of 21 October 2021 illustrates the hardest one. Bitcoin traded at $8,200 for 13 seconds, and every feed covering that venue recorded the number as a genuine low.

Gaps, zero-volume bars and forward fills

Two providers handle an empty interval differently. One emits no row, the other emits a bar carrying the previous close into all four price fields with zero volume. Code assuming one bar per interval misaligns the moment either series has a gap. Forward-filled closes cause more damage than gaps, because a flat synthetic candle resembles a real quiet period and suppresses measured volatility. Delisted pairs create the third variant. An OHLCV dataset assembled from currently-listed markets omits every token that failed, which inflates any return computed across the set. The remedy in each case is explicit: build a continuous index first, then decide deliberately what fills it.

Bad wicks and the Binance.US flash crash

On 21 October 2021 at 11:34 UTC, bitcoin fell from roughly $65,760 to $8,200 on Binance.US. The 87% drop reversed inside the same minute (CoinDesk, 2021; The Block, 2021). Binance.US attributed the move to a bug in an institutional client's trading algorithm. Arcane Research measured about 550 BTC trading during the 13-second window, against a typical 0.74 BTC of sell volume per four-hour period on that venue. Other exchanges fell around $1,000. The minute bar recorded a low of $8,200 and the daily bar inherited it. The OHLCV data was accurate. The price was not a market price, and a strategy backtested on that wick would have bought a fill that never existed.

Data current as of September 2026.

Most of these failures enter an OHLCV dataset at the moment of the API call.

How Do You Pull OHLCV From an API Without Corrupting It?

Three mechanical rules prevent most self-inflicted OHLCV bugs. Page past the per-request ceiling using the last close time. Keep every timestamp in UTC, and never treat the newest bar as final. Each rule maps to a documented limit rather than a preference.

Endpoint limits and pagination

Every OHLCV endpoint caps rows per request. Binance returns 500 klines by default and 1,000 at maximum, across intervals from one second to one month (Binance Spot API docs, 2026). Longer histories require pagination. The safe cursor is the last candle's close time plus one millisecond, because advancing by the interval reintroduces the boundary trade. CoinPaprika caps historical requests at 250 candles and restricts free-tier OHLCV history to the trailing 24 hours, with deeper archives on paid plans. Kraken publishes complete history as downloadable CSV files at eight fixed intervals, which removes pagination from the problem. Coverage differs as much as the limits: CoinPaprika tracks more than 12,000 coins across over 350 exchanges.

The candle that is still forming

The newest bar in any live OHLCV feed is incomplete. CoinPaprika's today endpoint makes this visible. A request at 11:53 UTC returned a candle stamped time_close 2026-09-10T11:53:00Z, 12 hours short of a full day (CoinPaprika API, 2026-09-10). Its high, low, close and volume all continue to move. Indicators recalculated on a forming bar flip signals repeatedly before the interval ends, which is the mechanism behind repainting. Backtests including the current bar leak information that was unavailable at decision time. The rule is to drop the final row from any live pull, or gate on the interval end before treating the bar as data.

Clean OHLCV answers a defined set of questions, and the last step is knowing which questions fall outside that set.

When Should You Move Below Candles to Trades or the Order Book?

OHLCV answers one question well: what happened to price and volume across a period. Once the question becomes whether a fill was achievable at a given price, the honest answer needs the trade tape or the order book. No increase in candle resolution substitutes for either.

What each data layer answers

Four layers sit beneath a chart. Order book data shows resting bids and offers at every price ↗, which is what depth, spread and slippage estimates require. Quote data thins that to the best bid and offer as they update, without the depth behind them. Trade data records every execution in sequence, with price, size and aggressor side. OHLCV compresses those trades into fixed intervals. Each layer answers a different question, and each costs one to three orders of magnitude more storage than the layer above it.

When candles stop being enough

Charting, trend analysis, indicator calculation and low-frequency backtesting run correctly on OHLCV data. Execution simulation, slippage modelling, liquidity analysis and any strategy holding positions for seconds do not. The dividing line is whether profit depends on the price available at the moment of the trade. Candle resolution offers no path across that line. A one-second bar still reports four prices and a total, with no record of the size resting on either side. Candles report a price that existed. Order books report a price that was reachable.

FAQ

References / Sources