Candle Data Does Not Tell the Whole Market

Candles look objective: open, high, low, close. But a candle is not the market. It is a summary of transactions from one source, during one interval, using one aggregation rule.

Two exchanges can produce different candles. One may have deeper liquidity while another receives a large trade seconds later. Both are valid, but neither represents every executable price.

OHLC also hides the order of events. Price may fall first and recover, or rise first and collapse. The final candle can look identical while execution risk is different.

Volume is not automatically healthy demand. It can include one group’s activity, wash trading, or position transfers that need order-book context.

Timeframe changes the story. A one-minute candle can be noisy while a daily candle looks calm. A bot must know whether the latest candle is closed; treating an unfinished candle as final creates signals that disappear later.

Use candles to form hypotheses, not proof. Validate the source, liquidity, spread, execution cost, and out-of-sample behavior.

Data is not neutral

Providers may fill empty candles, normalize timezones, filter anomalies, or rename symbols. Document those transformations. A backtest using only surviving assets has survivorship bias; include spread, fees, funding, withdrawals, and latency. A candle is an aggregation, not tick order. Ensure the latest candle is closed and test out of sample.

Candles hide the price path

OHLC does not show whether the high happened before the low, how long price stayed at a level, or available liquidity. Identical candles can contain different intrabar paths. A strategy may assume the target was reached first when the stop actually occurred earlier.

Volume needs context. Spot and futures differ; large activity may come from liquidations or wash trading. Read volume with spread, depth, open interest, trade size, and source.

The data pipeline can change a strategy. Check timezones, missing candles, duplicate trades, symbol changes, rounding, and unfinished candles. Store provider metadata and validate high, low, open, and close before backtesting.

Include fees, slippage, funding, latency, and delisted assets to reduce survivorship bias. Separate development data from evaluation data. Repeatedly tuning parameters against all history is memorizing the past.

Data-quality checklist

Sample normal days and volatile days. Compare timestamps, OHLC, volume, and trade counts with the source exchange. Mark synthetic candles and missing intervals. Freeze the evaluation dataset so it does not change while parameters are tuned. Version transformation code with backtest results. If results cannot be reproduced from raw data, the profit figure is not evidence—only a convincing-looking output.

Sources

• https://www.cmegroup.com/education/courses/technical-analysis/understanding-candlesticks.html
• https://www.investopedia.com/terms/o/ohlc.asp
• https://www.tradingview.com/support/solutions/43000561891-what-are-bars-and-candles/