Tick-by-tick EA versus once-per-bar logic: same idea, different strategy

[Shadow Trader]

An EA that evaluates conditions on every tick behaves very differently from one that evaluates only when a new bar opens. The code might look nearly identical, but the strategy is not the same.

Tick-based logic reacts immediately. It also reacts to noise: a quick spike that reverses before the bar closes can trigger an entry that wouldn't exist on a closed-bar basis. Backtests that model every tick capture some of this, while backtests on bar open prices ignore it.

Bar-based logic is calmer and easier to test, but it enters late. On a one-minute chart, an entry at the start of the next bar can be 59 seconds behind the signal. For a scalper aiming for a few pips, that delay is a big share of the whole move.

A third option is evaluating on every tick but requiring a condition to persist for a set number of ticks or seconds. It filters noise without waiting for the bar.

An often missed point is that live behavior and backtested behavior can diverge here. The strategy tester generates ticks from bar data unless real tick data is supplied, and the sequence inside a bar is an approximation.

It means that a tick-based EA can look great in a backtest and fail in live trading, or the reverse, depending on how the test modeled the movement within each bar.

In the strategy tester, the "every tick" mode simulates intrabar movement from bar data, using an assumed pattern, which means tick-based logic is tested against synthetic ticks. The "open prices only" mode skips intrabar behavior altogether and is only valid for logic that acts on bar open.

Choosing the mode that matches the code is basic, yet many tests use the wrong one.

Which logic style has held up best in live trading, tick or bar?

[Amanhasnoname]

Bar logic has held up much better for me, though I resisted it for a long time. My first EA was a simple moving average cross that checked every tick, and in testing on generated ticks it looked fantastic. Live, it kept entering on crosses that existed for ten seconds in the middle of a bar and then disappeared, and those trades were mostly losers.

Switching to evaluation at the open of each new bar made the live results look almost identical to the test, which was a relief, but the late entries cost me around a pip on average on the one minute chart. For a system aiming for six or seven pips, that was a real chunk of the edge.

The middle option you describe is where I ended up. The EA checks on every tick but needs the condition to hold for fifteen seconds before it acts. I had to test it on real tick data rather than the platform's simulated ticks, because the generated data made the persistence filter look far better than it really was.