Tick data

A tick is a single quote or trade update, one moment the market actually printed a price. A one-minute candle can compress hundreds of them into four numbers. A one-hour candle, thousands. Whatever gets thrown out in that compression is gone for good, and the backtest starts already blind to it — before a single rule has fired, before cost even enters the picture.

That matters most at the edges of the bar, because a candle keeps no record of the spread. It shows one price, not the two that were actually tradable at any given instant: the bid you'd have sold into and the ask you'd have bought at, captured properly only in the bid/ask spread. Sample once per bar and that gap simply disappears from the data.

Backtest on candle closes and the strategy trades at a phantom midprice that was never a quote anyone could deal at. Every entry looks a touch better than it would live. Every exit looks a touch worse than reality demands. Stack that across thousands of trades and a strategy's equity curve ends up flattering itself with fills nobody could have gotten.

For FX we run genuine tick data with real bid/ask so spread and slippage get measured instead of guessed, and we only test timeframes of 30 minutes and up — sub-15-minute FX is cost-fatal before a strategy ever gets a fair shot. For futures we build 13 years of history off CME data, stitched into continuous contracts so a strategy trades one clean price series instead of a pile of expiring months bolted together.

None of this is academic. A system that clears on midprice can be underwater the moment you charge it what a real trade actually costs to enter and exit. Cheap data produces cheap answers dressed up as rigor. If the cost model behind a backtest is a guess, the verdict on the strategy is a guess too, no matter how clean the equity curve looks.