Backtestify
Sign in
All articles
Tick Data BacktestingNovember 3, 202614 min read

Tick Data Backtesting: When It Matters and How to Do It

Learn what tick data backtesting is, when it beats candle data, how to clean tick data, model costs and fills, and avoid look-ahead bias before going live.

Tick Data Backtesting: When It Matters and How to Do It

Tick data backtesting means replaying a strategy against the individual trades and quote updates the market actually produced, instead of against summarized candles. It matters most when your strategy's result depends on what happened inside a bar: scalping, high-frequency logic, tight stops and targets, or execution timing. For slower strategies, the quality of your cost assumptions and your validation usually matters more than the data resolution. This guide covers what tick data is, when it earns its cost, how to prepare it, and how to model fills so the result survives contact with live trading.

TL;DR

  • Tick data records individual trades and quote updates. Candles compress them and lose the order of events inside each bar.
  • Research on candle-based testing finds that tick data cuts backtest errors mostly in high-frequency settings, and the gains shrink for slower strategies.
  • Clean, correctly timestamped data is the first defense against look-ahead bias. Spreads, slippage, commissions and latency come next.
  • No backtest can promise a live result. Compare backtest and live performance over time to see how large the gap really is.
  • Match data resolution to your strategy's holding period, then verify the result with out-of-sample checks.

Table of Contents

What tick data is and how it differs from candles

A candle (or bar) summarizes everything that happened in a fixed window into four prices, open, high, low and close, plus volume. Tick data is the raw material underneath: each trade execution and each change to the best bid or ask, stamped with a time that is often precise to the millisecond or finer.

The difference is sequence. A one-minute candle tells you the price touched both 100.20 and 99.80 during the minute. It does not tell you which came first. If your strategy has a stop at 99.85 and a target at 100.15, the candle cannot say whether you were stopped out or paid. Tick data can.

Two flavors are worth separating:

  • Trade ticks record executed transactions: price, size, time.
  • Quote ticks record changes to the best bid and offer. They let you model the spread you would have paid at the moment you traded.

Deeper feeds, often called Level 2 or Level 3, add market depth and give a fuller picture of liquidity, according to an LSEG expert talk on tick history. Most retail strategies do not need depth, but a strategy that trades size relative to what is displayed does.

When tick data gives you a real advantage

More data is not automatically better. It is heavier to store, slower to run and easier to misuse. The question is whether the information inside the bar changes your trade decisions.

An academic thesis from Aalto University finds that candle data loses price movement within candle periods, which can make candle-based backtests less accurate. It also reports that tick-data backtesting reduces errors compared with candle-based testing in high-frequency contexts, while the gains diminish for lower-frequency algorithms. That is a useful rule of thumb: the shorter your holding period, the more the inside of the bar matters.

Cases where tick data usually pays off

  • Scalping and very short-term intraday systems. Trades last seconds or a few minutes, and a single bar can contain the entry, the stop and the target.
  • Tight stop and target pairs. When both sit within one bar's range, only the tick sequence resolves which was hit first.
  • Strategies sensitive to spread. If your average gain per trade is a few ticks, the spread at the moment of entry decides whether the edge exists.
  • Execution research. LSEG's discussion notes that high-frequency data helps even daily strategies, because you can simulate execution at different times of day instead of assuming a fill at the closing price.

Cases where bars are usually enough

Swing and position strategies that enter on a daily close and hold for days rarely hinge on intrabar order. Their risks sit elsewhere: too short a data history, parameter tuning that fits noise, and unrealistic costs. If that is your style, our guides on swing trading backtests and how much historical data you need address the problems that usually bite first.

A middle path is common: develop and screen ideas on bars, then re-test the survivors at a finer resolution before risking money. Our piece on intraday backtesting covers how to structure that for day trading.

Sourcing and cleaning tick data

The LSEG talk is blunt about the baseline: historical tick data should be properly timestamped and cleaned before it goes into a backtest. In practice that means working through a checklist before you trust a single result.

A pre-flight checklist for tick data

  1. Confirm what the feed contains. Trades only, or trades plus quotes? Consolidated across venues, or from one exchange? A single-venue feed can miss prints and misstate the best price.
  2. Check timestamps. Know the time zone, the clock source and the precision. Mixed time zones or daylight saving shifts silently misalign data from different sources.
  3. Look for gaps. Compare tick counts per session against what you expect. A flat stretch in a liquid market is usually missing data, not calm.
  4. Filter bad prints. Isolated prices far from neighbors, zero or negative values and duplicated records distort stops and extremes. Decide your filtering rules in advance and apply them uniformly.
  5. Handle session boundaries and corporate actions. Roll dates for futures, splits and dividends for stocks, and holiday schedules all change what a continuous series means.
  6. Keep a raw copy. Store the untouched data separately so you can redo cleaning if you change your mind.

A practical test: pick three days, plot the ticks, and compare them with a chart from a second source. If the highs and lows disagree materially, find out why before running anything.

Free and cheap tick sources often differ from institutional feeds in coverage, depth and cleaning. That does not make them unusable, but it does mean you should state which data your result came from.

Avoiding look-ahead bias with tick data

Look-ahead bias means your test uses information that was not available at the moment of the decision. Tick data gives you more ways to commit it, not fewer, because finer resolution tempts you to act on a price the instant you see it.

LSEG's guidance is to ensure data is properly timestamped and cleaned to avoid look-ahead bias. Beyond that, these habits catch most leaks:

  • Separate signal time from order time. If a signal is computed from a tick at time T, the earliest realistic fill is after T plus your latency, never at T.
  • Never use a bar's final high or low inside that bar. A rule such as "buy if price breaks the high of the current candle" is only knowable after the candle closes.
  • Compute indicators only from completed data. A moving average updated tick by tick must use only ticks that have already occurred.
  • Watch pattern definitions that need future bars to confirm. Swing highs, order blocks and similar structures are often labeled with hindsight. We walk through four quick tests for look-ahead bias, and the same trap shows up in smart money concept backtests and order block backtests.

Overfitting is the second half of the problem. Tick-level tests run thousands of trades, which can make a fitted result look statistically solid when it is not. Fix your rules and parameter ranges before you look at results, hold out a later period you never tune on, and prefer parameter zones that work broadly over a single sharp peak.

Modeling spreads, slippage, commissions and latency

This is where tick data earns its keep, and where many tick backtests still fail. Having precise prices does not help if you assume you can always trade at them.

Spread

Use quote ticks when you have them. Buy at the ask and sell at the bid, rather than trading at a midpoint or last trade. For a strategy that aims at a few ticks per trade, this single change can erase a backtested profit.

Slippage

Slippage is the gap between the price you expected and the price you got. Model it as a function of conditions: wider around news and at the open, narrower in the middle of liquid sessions, larger for bigger orders. A fixed number is a start, but a variable one is closer to reality. If you test on minute bars instead, a fill formula can approximate the same effect, as shown in this minute bar slippage formula.

Commissions and fees

Include every per-trade cost: commission, exchange and regulatory fees, and any data or platform costs you want in the return calculation. The LSEG discussion stresses using accurate trading-cost assumptions in a tick backtest.

Latency and fill logic

Between your decision and the exchange's acknowledgment, time passes and the book moves. Add a delay to every order, based on your real setup, and fill at what the market showed after that delay. For limit orders, be conservative: being touched is not the same as being filled. Queue position matters, and a trade at your price does not guarantee your order was executed.

Latency is also a market-wide cost. A 2021 Bank for International Settlements working paper estimates that latency arbitrage imposes roughly a 0.5 basis point tax on trading, and that the modal race lasts 5 to 10 microseconds. A separate BIS study on dark pools found that high-frequency firms were on the profitable side of trades 96 to 99 percent of the time when reference prices were stale. You are unlikely to compete in those races as a retail trader, which is the point: if your strategy's edge lives at that speed, a backtest that ignores latency is describing trades you could not have made. For a closer look at how message-level data exposes these races, see this explainer on microsecond latency arbitrage.

Latency and fill logic

Validating the result: is the backtest robust?

A finished backtest is a hypothesis. These checks decide how much weight it deserves.

Read the right metrics together

Look at win rate, profit factor and maximum drawdown as a set. A high win rate with a poor profit factor means small wins and occasional large losses. A good profit factor with a deep drawdown may be unbearable to trade. Also check the number of trades: a few dozen is anecdote, not evidence.

Split the data

Develop on one period, then test on a later period without changing the rules. Rolling the split forward across several windows (walk-forward testing) shows whether the edge persists across regimes or belongs to one market phase.

Stress the assumptions

Double your slippage. Add a few hundred milliseconds of latency. Raise commissions. If the strategy collapses under modest pessimism, it has no margin of safety.

Compare with live results

The Aalto thesis suggests assessing robustness by comparing backtest results with live results and measuring backtest error over time, including cumulative error. A small live or paper-traded sample, tracked against what the backtest would have done on the same days, tells you how much optimism is built into your assumptions. Our step-by-step backtesting guide and the piece on evidence-based trading lay out a full validation routine.

Can a backtest predict the next move?

No. LSEG states plainly that backtesting cannot tell you with certainty what a strategy's future performance will be, and past performance is no guarantee of future results. Tick data makes the replay of the past more faithful. It does not make the future repeat it.

What a rigorous backtest does give you is a distribution: how often the rules won, how large the losses got, how long the bad stretches lasted. That lets you size positions and decide in advance what drawdown means "the idea is broken" rather than "normal variance". Forecasting tools that express outcomes as probabilities and confidence levels, rather than single predictions, fit that mindset better than any promise of certainty.

Putting it into practice

A workable process for most traders:

  1. Write the rules in unambiguous terms: entry, exit, stop, target, sizing, session.
  2. Run a first pass on bar data to see whether the idea has any signal across a long history.
  3. If the logic depends on intrabar order, short holding times or tight stops, re-run at tick or quote resolution on a representative period.
  4. Apply realistic spreads, slippage, commissions and latency.
  5. Validate on data you did not tune on.
  6. Paper trade or trade small, and track the gap against the backtest.

If you are testing a setup you saw from a trading creator, the same process applies. Publicly shown results are rarely accompanied by full rules or cost assumptions, so reproducing them under your own explicit rules is the honest test. The Published Strategy Library shows how reports on well-known setups look when the rules are written down and the results are shown whether they win or lose.

Test your strategy with Backtestify

Backtestify lets you describe a strategy in plain English, run it against real historical market data, and see the metrics that decide whether it deserves capital: win rate, profit factor, maximum drawdown and a trade-by-trade breakdown. You can then improve the rules and compare the original and improved versions side by side on recent data, or use the forecast tool to see probability and confidence statistics for the next price movement. Results can be exported in TradingView Pine Script format. The methodology page explains how a strategy is simulated, so you can check our assumptions against yours.

Backtestify

You can start on the Free plan. Pro, with unlimited backtesting, improvement and forecasting, costs $29 per month or $190 per year; see pricing for what each plan includes.

FAQ

Is tick data necessary for every backtest?

No. Tick data is most valuable for high-frequency, scalping and tight-stop strategies, where the order of events inside a bar changes the outcome. Research on candle versus tick testing finds the error reduction shrinks as strategy frequency falls. For swing or position strategies, realistic costs and out-of-sample validation usually matter more.

What is the difference between tick data and one-minute bars?

A one-minute bar keeps only the open, high, low, close and volume for that minute. Tick data keeps every trade or quote update in sequence, so you can tell which price came first. That sequence decides whether a stop or a target was hit first inside a bar.

How do I avoid look-ahead bias in a tick-data backtest?

Use clean, correctly timestamped data. Compute signals only from events that have already happened, and fill orders after a realistic delay, never at the same tick that triggered the signal. Be careful with patterns that need later prices to confirm, since they are often labeled with hindsight.

Will a tick-data backtest guarantee live results?

No. Backtesting cannot tell with certainty how a strategy will perform in the future. A finer-grained test reduces one source of error, but changing markets, execution differences and overfitting remain. Compare live and backtested results over time to measure the gap.

Sources

Recommended