Real ticks vs 1-minute OHLC: which EAs actually change, and by how much

Everyone repeats "use real ticks", but almost nobody says what it costs you to ignore that advice. I re-ran a set of Expert Advisors under both tick models on the same symbol, the same period and the same broker history. The answer is very uneven: for some EAs the two models agree closely, and for others the result is not merely different, it is the opposite sign.

A gold grid scalper with a take-profit of about 1.5 USD per position looked excellent on the 1 minute OHLC model: profit factor 2.46. Re-run on "every tick based on real ticks", same code, same period: profit factor 0.488, and the account ends about 65% down. What convinced me this was an artefact rather than noise is that the number of trades was nearly identical in both runs. The model was not finding different opportunities, it was filling the same ones at prices that never existed. When the take-profit is thinner than the intrabar path, OHLC interpolation hands you fills for free.

The bias is not always optimistic, and this is the part people miss. A gold breakout EA with an ATR stop-loss and an ATR take-profit showed a complete blow-up under the 1 minute OHLC model. On real ticks, across two different brokers' histories, the same EA was profitable with a profit factor around 1.3. The OHLC model has to guess the order in which the high and the low of the minute were touched. When the stop and the target both sit inside the same minute's range, that guess decides the trade - and guessing wrong often enough makes a viable system look dead.

A Donchian breakout EA - entry only on the close of a completed bar, a hard stop several ATR away, an ATR trailing stop, no fixed take-profit, one position at a time - produced essentially the same picture under both models. USDJPY H1 on real ticks, January 2017 to May 2026: profit factor 1.50, maximum equity drawdown 10.3%, 631 trades, win rate 34.5%. Nothing in that logic depends on where price travelled inside a bar, so there is nothing for the tick model to get wrong.

The tighter the exit relative to the intrabar range, the more the tick model - not the market - decides your backtest. Thin take-profit, grid, scalping, anything that can be opened and closed inside a single minute: real ticks are mandatory, and a modeled-tick result is worthless. Entries and exits on bar closes with wide ATR-based stops: the model barely matters, and OHLC is a legitimate way to run long optimisations quickly. Anything where the stop and the target sit inside the same one-minute bar: treat both models as unreliable and inspect the fills by hand.

Two practical consequences. First, a published backtest that does not state its modelling method is not a result, it is a picture - and the omission tends to correlate with which model was chosen. Second, if you are optimising a bar-close system, you can honestly sweep on OHLC and confirm the winner on real ticks; you do not need every-tick runs during the search.

If you would rather reproduce case 3 than take my word for it, that EA is free and its source is published: Donchian Trend Engine. Run it once with each tick model and compare.

I would genuinely like a counter-example: an EA without a thin take-profit where the two models still diverge sharply. That would break the rule above, and I would rather find out now than after funding it.