Details
What each one looks like
Here are the first three trades of 1 November 2026 on BTC-USDT, from our free trades sample:
| symbol | trade_id | timestamp | price | size | notional | aggressor_side |
|---|---|---|---|---|---|---|
BTC-USDT |
6809266910 | 2026-11-01T00:00:00.447286Z | 77582.02 | 0.00105 | 81.46 | sell |
BTC-USDT |
6809266911 | 2026-11-01T00:00:00.452197Z | 77582.02 | 0.00217 | 168.35 | sell |
BTC-USDT |
6809266912 | 2026-11-01T00:00:00.740755Z | 77582.03 | 0.00025 | 19.4 | buy |
Source: Binance
The file holds 500 trades, and all of them happened in the first 29.5 seconds of the day. In that half minute buyers took 1.391 BTC and sellers hit 1.401 BTC, almost perfectly balanced.
And here is a 1-minute bar from the bars sample, the last minute of 31 October:
| symbol | interval | timestamp | open | high | low | close | volume | trade_count |
|---|---|---|---|---|---|---|---|---|
BTC-USDT |
1m | 2026-10-31T23:59:00Z | 77580.28 | 77585.09 | 77580.28 | 77582.02 | 1.62577 | 512 |
Source: Binance. Columns quote_volume, vwap and taker_buy_volume omitted here.
That one row stands for 512 trades. You keep the range and the volume. What disappears is the sequence: whether the high came before the low, how big each trade was, and who was aggressive at each moment. The bar file does keep taker_buy_volume, which recovers part of that last point.
The size difference
The 500 trades above cover less than 30 seconds. At that pace, one busy day of BTC-USDT trades is millions of rows. The same day in 1-minute bars is 1,440 rows. For a US stock, a year of 1-minute regular-session bars is about 98,000 rows; a year of trades for an active name can be hundreds of millions.
Side by side
Turn ticks into bars and compare
"""Example: build 1-second bars from ticks and measure what bars hide."""
import pandas as pd
t = pd.read_csv("btc-usdt-trades.csv", parse_dates=["timestamp"]).set_index("timestamp")
bars = t["price"].resample("1s").ohlc().dropna()
bars["volume"] = t["size"].resample("1s").sum()
bars["buy_volume"] = t.loc[t["aggressor_side"] == "buy", "size"].resample("1s").sum()
bars["trades"] = t["price"].resample("1s").count()
print(len(t), "trades ->", len(bars), "one-second bars")
print(bars.head())
first_extreme = t["price"].resample("1s").apply(
lambda p: "high first" if p.idxmax() < p.idxmin() else "low first")
print(first_extreme.value_counts())
The last block shows, for every second, whether the high or the low came first. That is exactly the information a bar backtest has to guess. If your strategy's stop and target often sit inside a single bar's range, that guess decides the result and you need ticks, or at least much smaller bars.
A simple rule for choosing
- Hold for days, stops wider than a day's range: daily bars are enough.
- Intraday, holding minutes to hours, stops several bars wide: 1-minute bars work, with a realistic spread added.
- Scalping, tight stops, queue position, or measuring slippage: you need ticks, and for spread-sensitive work you need quotes as well as trades.
Watch out
Common mistakes
Assuming the bar's close is a tradable price. It is the last trade in the interval. On thin markets it may be an old, small trade far from the current quote.
Using trades when you need quotes. Trade ticks tell you where deals happened, not the bid and ask you would have faced. For fill modelling in stocks and FX you want quote data.
Building bars with the wrong labels. Decide whether a timestamp marks the bar's start or end, and check your vendor's choice. Ours marks the start.
Buying ticks you won't use. If your strategy holds for a week, tick data adds cost and storage without changing the result.
Details
Get the data
Ticks and bars in the same schema, so you can move between them. Free samples:
- Download BTC-USDT trades (500 ticks)
- Download BTC-USDT 1-minute bars (500 bars)
- Source: Binance
- Stock trades: stock tick data, available in early access, suggested price $15 per symbol-year. A stock tick sample is available on request.
FAQ
Frequently asked questions
What is the difference between tick data and bar data?
Tick data has one row per trade (or per quote update), with exact time, price and size. Bar data groups those ticks into fixed intervals and keeps only open, high, low, close and volume. Ticks keep the order of events; bars do not.
Do I need tick data for backtesting?
Only if your strategy's outcome depends on what happens inside a bar, such as tight stops, scalping or slippage studies. For daily or multi-hour strategies, bars give the same answer with far less data.
How big is tick data?
Much bigger than bars. An active crypto pair or US stock can produce millions of trades a day, compared with 1,440 one-minute bars for a 24-hour market. Store ticks in Parquet or compressed CSV.
Can I make bars from tick data?
Yes. Resample the price with first, max, min and last, and sum the sizes. The reverse is impossible, which is one reason to keep ticks if you might need finer detail later.
Keep reading