Intrabar order is the thing OHLC throws away

[Shadow Trader]

A one-minute candle gives four numbers: open, high, low, close. What it can't say is the order in which the high and the low occurred. Was the high first, then a fall to the low? Or the low first, then a rise to the high?

For a scalper with a 3-pip target and a 3-pip stop in the same bar, that order decides whether the trade won or lost. Both levels were touched. Which one came first is the whole question.

Backtests on bar data have to guess. Most testers assume a convention: for a bullish bar, the low came first, then the high. Or they use a pessimistic assumption, in which the stop is always hit first. The results depend heavily on the assumption, and neither matches reality consistently.

Tick data resolves it. With every price change recorded in sequence, the sequence is known. A strategy with tight targets can show completely different profits on tick data compared with bars.

The gap grows as the target shrinks. A swing trader aiming for 100 pips can ignore the problem. A scalper aiming for 2 or 3 pips can't.

It's the reason why any scalping backtest using only bar data should be treated as a rough sketch. The equity curve might be correct in shape, but the details are more or less fiction.

A quick way to see the damage: take a strategy with a 2 pip target and a 2 pip stop. On bar data with a pessimistic convention, any bar that touches both levels counts as a loss. With an optimistic convention, it counts as a win. On a busy pair, 15 to 25 percent of bars can touch both levels, and the results can swing from clearly profitable to clearly unprofitable depending on the convention.

That swing is the size of the uncertainty bar data leaves. Tick data narrows it, though it doesn't remove other issues such as spread and slippage.

How big was the difference when a scalping test was repeated on tick data after being run on bars?

[Amanhasnoname]

I learned this the expensive way with a tight GBPUSD breakout system I'd built on one minute bars. The tester said it was nicely profitable with a two and a half pip target and a matching stop. When I reran it on tick data I'd recorded myself over a couple of months, about a fifth of the trades that had been counted as winners turned into losses, because the stop had been touched first inside the bar.

The result went from a decent edge to roughly break even before costs and clearly negative after them. What surprised me was how concentrated the damage was. Almost all of the flipped trades happened in the first fifteen minutes of London and in the minutes after US data, where bars were big and both levels got hit.

Since then I won't trust a bar based test for anything with a target under about ten pips. If I only have bars, I run it twice, once with the pessimistic assumption and once optimistic, and if the strategy only works under the optimistic one, I drop it.