Tick data is the record of every individual price change on a symbol — each one timestamped, with its own bid and ask — and it is what MT5 replays in its 'Every tick based on real ticks' model instead of inventing a path between bars.
also: tick history, real tick data
Updated · Reviewed
Bar data keeps four prices per candle: open, high, low and close. Tick data keeps everything between them — every price change, in order, carrying the spread that stood at that moment. The gap between the two only decides anything when a trade opens and closes inside a bar, which is exactly where most EA marketing lives.
A backtest can only trade the prices you hand it. Bar data hands the tester four per candle and lets it invent the route between them, so any strategy that enters and exits inside the bar answers for a path nobody recorded. Tick data swaps the invention for the record. Our own runs then add an uncomfortable second finding: where you take the ticks from moves the result far more than switching from invented ticks to recorded ones.
One EA, one window, three runs. TidewellSlack trades GBP/JPY on M15 from 2019-01-01, and we ran it three ways to prise apart the two things listings routinely fold together: where the ticks come from, and whether the tester replays them or invents them.
Two of the three runs fail our own gate, and the survivor is the one on generated ticks. That is why the listing publishes its symbol and its model number rather than the adjective 'real'.
Calculation Source swap: −0.22 of profit factor and +8.26 points of drawdown. The real-tick upgrade on top of it: −0.02 and +0.02.
Result The data source moved the result eleven times as far as the tick model did
Read tick data against holding time first and against the source second. The finer your target relative to the bar, the more of the outcome the path decided — but the source decides how much of that outcome travels anywhere else.
| Range | What it means |
|---|---|
| Intraday or scalping EA, real ticks, vendor named | The defensible case. The path is recorded and a stranger can rebuild the run from the listing alone. |
| Multi-day swing system on bar-generated ticks | Usually adequate. One bar's internal route rarely decides a trade held for a week. |
| Real ticks, but from a single source | Half an answer. Our GBP/JPY run passed on one source and failed on another with identical rules and an identical window. |
| 'Real ticks' with no vendor and no model number | Unverifiable, and the two missing fields are precisely the two that decide the figure. |
Eleven of the 23 runs published here use Model 4 and twelve use Model 0, and every Model 4 run carries a _DUKA suffix on its symbol name because imported ticks can only live on a custom symbol. Ten of those eleven manifests still name the broker's M1 history in their price-source field, a leftover from the lineage they migrated out of. Read the model number and the symbol; the prose around them ages faster.
The model is one dropdown line and the source is a whole different market's record. On the GBP/JPY run above, changing the source cost 0.22 of profit factor and doubled the drawdown, while the real-tick upgrade on the same source cost 0.02. A listing that names its model but not its vendor has told you the smaller half.
The percentage measures how completely the history covered the window, not which model produced it. The Model 0 shadow run behind KestrelHover printed 100% — the same figure its real-tick sibling printed. A number that both models can reach cannot distinguish between them.
On Model 4 the tester fills from the tick file's own ask minus bid and disregards the field entirely; running one USD/JPY test at Spread=10 and again at Spread=100 returned identical tick-by-tick spreads. The setting changes nothing in the run and quietly changes what the author believes the run measured.
A test from 2015 on a source whose ticks open in 2019 interpolates the early years, replays the later ones, and reports the two as a single result. The Ticks tab states the range it holds; the tester never warns you that your window overshot it.
They close the price half and leave the execution half untouched. Latency, requotes, partial fills and the broker's policy stay outside the simulation, so a real-tick run remains an optimistic account of what the fills would have been — just an optimistic account built on a path that genuinely happened.
| TidewellSlack — GBP/JPY, M15, from 2019-01-01 | Trades | Profit factor | Max drawdown | History quality | Our gate |
|---|---|---|---|---|---|
| Broker M1 history, Model 0 (published) | 303 | 1.35 | 9.92% | 99% | PASS |
| Imported ticks, Model 0 | 351 | 1.13 | 18.18% | 99% | FAIL |
| Imported ticks, Model 4 real ticks | 345 | 1.11 | 18.20% | 100% | FAIL |
Row one to row two changes the data and nothing else: same rules, same window, same tester model. The profit factor drops 0.22 and the drawdown roughly doubles. Row two to row three changes the model and nothing else, upgrading generated ticks to the recorded ones from the same archive, and the profit factor drops 0.02 while the drawdown moves two hundredths of a point. Both imported-tick rows fail our gate; the survivor is the run on generated ticks.
That ordering is not universal, which is the honest part. KestrelHover went the other way: 291 trades and a 1.33 profit factor on real ticks, against 292 trades and 1.30 on the broker’s generated ticks, and both runs cleared. Sometimes the source is the whole story and sometimes it is noise — and you only find out by running it twice. Read TidewellSlack’s listing next to that one and the point makes itself.
Eleven of the 23 runs published here use Model 4 and twelve use Model 0. Every
single Model 4 run trades a symbol whose name ends in _DUKA, because imported ticks
cannot live on the broker’s own symbol — you create a custom symbol in the Symbols
window and import the archive into it. Not one broker-native symbol in this catalogue
ever reached the real-tick model, which is the practical reason so many published
backtests do not use it: the mode is free, the inventory behind it is not.
That also breaks the usual shortcut of trusting a percentage. The Model 0 shadow run behind KestrelHover printed 100% history quality, the same figure its real-tick sibling printed, so the number cannot tell the two apart. Modelling quality answers a different question — how completely the history covered the window — and answers it for both models.
Tick data’s second gift is the recorded bid and ask, and it is the one most often thrown away. On Model 4 the tester fills from each tick’s own ask minus bid and ignores the spread you configured: running one USD/JPY test at Spread=10 and again at Spread=100 produced identical tick-by-tick spreads, 227 distinct values ranging from 10 to 350 points. The field looks like a control and behaves like a comment.
Almost nobody measures the resulting cost either. Exactly one of the 23 runs here states a measured spread average — 19.14 points, recorded inside the tester — and that run uses Model 0, not real ticks. So the honest reading of most real-tick claims, including several of our own, is that the path was recorded and the cost of walking it was never totalled.