Working With Historical Market Data

How do I get historical candle data from CoinDock?

Through the public candles API — no account required:

GET /api/v1/markets/{symbol}/candles?interval=1h

Intervals: 1m, 5m, 15m, 30m, 1h, 4h, 1d, 1w.

Each candle carries:

The last two are the ones most APIs omit, and they are what let you tell a real market from a few transactions.

Why does trade count matter so much for historical work?

Because a candle is a summary, and a summary of four trades is not data you can analyse.

Two candles can be numerically identical while one aggregates 3,000 trades and the other four. Any statistic computed across a dataset containing many low-count candles is measuring artefacts, not market behaviour.

Filter or flag low-trades_count candles before any analysis. This single step invalidates a large share of amateur crypto backtests that skip it.

What should I do about unclosed candles?

Exclude them, or handle them explicitly.

The most recent candle is usually still forming: its close is the last trade so far, and its high and low can extend. Including it in analysis mixes a partial observation with complete ones — and in a live strategy it produces the classic error of "signals" that disappear when the candle closes.

Filter on is_closed.

What is survivorship bias here, and why is crypto worse?

Analysing only assets that still exist, and concluding things that do not hold for the population you would actually have chosen from.

Crypto is unusually affected: enormous numbers of tokens have been delisted, abandoned, or gone to zero. A study of tokens listed today has silently excluded most of what a trader in the past could have bought.

Practical consequence: any historical result derived only from currently listed tokens overstates outcomes, sometimes dramatically.

What else makes crypto backtests unreliable?

• Thin-market candles — covered above; the largest single problem.
• Fabricated volume — wash trading contaminates volume-based rules. See volume FAQ.
• No slippage modelling — assuming fills at the close ignores that your order moves the price. On thin books this alone can turn a profitable backtest into a losing strategy. See understanding slippage.
• No fee modelling — maker/taker fees, and gas on DEXs.
• Look-ahead bias — using information not available at the decision point, of which unclosed candles are the most common instance.
• Overfitting — tuning parameters until historical results look good. If a rule works at 14 periods and fails at 13 and 15, you have fitted noise.
• Venue differences — candles differ between exchanges, so results may not transfer.

How much history is enough?

There is no threshold to quote, but two principles hold:

Enough regimes, not just enough rows. A year of data covering only rising prices tests one condition. More intervals of the same regime add rows, not evidence.

Out-of-sample data reserved from the start. Hold back a period before you begin and do not look at it until you have finalised the rule. Testing on data you have already explored is not testing.

Can I compare CoinDock candles with another exchange's?

Only carefully. Check that both use the same interval boundaries and time zone, the same timestamp convention (labelled by open or close time), and the same empty-interval handling — some venues omit candles, others carry the previous close forward.

Mismatches on any of these produce apparent disagreement where none exists. And venues legitimately differ anyway, because each aggregates only its own trades. See OHLC charts explained.

What is realistic to conclude from historical data?

Descriptive statements are well supported: how volatile a pair has been, how volume and trade counts have varied, how spreads have behaved, when activity clusters.

Predictive statements are much weaker, for all the reasons above. The gap between "this happened" and "this will happen" is where most analysis goes wrong.

CoinDock publishes market data, not forecasts. We make no claim that historical patterns predict future prices.

Related

• OHLC Charts Explained
• Charting FAQ
• Public API documentation

Educational content. Not trading advice. Historical performance does not indicate future results.