Realistic backtesting with historical bid & ask data.

Explained

What realistic backtesting means

A backtest is a claim about what a set of rules would have done. It is only as honest as the prices it fills at. Realistic backtesting means pricing each historical fill the way the market would actually have priced it — on the right side of the book, at a spread that existed at that moment — rather than at the midpoint of a chart or at a cost that was guessed once and applied to every trade.

It does not mean a backtest becomes a forecast, or that it matches a live account to the cent. It means one large, systematic source of optimism is measured instead of assumed. How much that matters depends on the strategy: for a swing trade with a wide stop, very little; for a scalp with a stop a few points away, it can be the difference between an edge and a loss.

Bid, ask and spread

At any moment a market has two prices. The bid is the best price someone will pay you, so it is what you receive when you sell. The ask is the best price someone will sell to you, so it is what you pay when you buy. The spread is the ask minus the bid — the cost of crossing from one side to the other, paid once on every round trip.

Most charts, including Sage's, draw the bid. That is why a long trade shows a small loss the moment it opens: it was bought at the ask and is valued at the bid. The spread is not a fee printed on a statement, so it is easy to forget — and a backtest that forgets it quietly hands every trade a discount it would never get live.

How spread affects entries, exits, stops and targets

Every order is either a buy or a sell, so every order trades on one side of the book. A long position is opened with a buy and closed with a sell; a short is the reverse. That decides which price each level is judged against:

The consequence is lopsided in a way that matters. On a bid chart, a short's stop can be hit while the drawn candle stops short of the line, because the ask above it reached the level. A short's target needs the drawn candle to travel a spread past the line before the ask gets there. A long pays the spread up front, at entry. None of this is visible if the backtest treats the chart line as the price.

Fixed spread vs historical spread

The usual fix is one fixed spread for the whole test. It is far better than none, and it is still wrong in a specific way: the quoted spread is not a constant. It changes through the day, it widens sharply around the daily rollover and around news, and it drifts over months and years as conditions change.

In Sage's stored gold (XAUUSD) data for 12 March 2025, the quoted spread had a median of 0.53, and eight minutes in ten sat between 0.45 and 0.64. At 20:59 UTC, around the daily rollover, it reached 2.00 — nearly four times the median. On 7 February 2024 the same instrument's median was 0.31. A single fixed number chosen for a test covering both days would be too generous on one and too harsh on the other, and wrong by the most at exactly the minutes when spread is widest.

The two are not rivals. Historical spread shows how cost varied; a fixed spread lets you model a particular broker or ask "what if it were always this bad?". Sage lets a session use either, and replaying the same dates both ways is a quick way to see how sensitive a result is to cost.

A worked example

Take a short on gold, sold at a bid of 2,912.00, with a stop 3.00 above and a target 6.00 below — 2R on the chart. Both exits are buy orders, so both trigger on the ask:

Spread charged        0.20 fixed    0.53 median    2.00 at rollover
Stop fires at bid     2,914.80      2,914.47       2,913.00
Bid rise to stop      2.80          2.47           1.00
Target fills at bid   2,905.80      2,905.47       2,904.00
Bid fall to target    6.20          6.53           8.00

Illustrative trade. Spreads: Sage's stored XAUUSD data, 12 March 2025.

The trade is still 1R lost or 2R won — the levels did not move. What moved is how far the drawn price has to travel to reach them. With the day's median spread the stop is about 18% closer and the target about 9% further away than the chart suggests; in the minutes around rollover, a 1.00 rise in the bid is enough to stop the trade out. A long is the mirror image: it pays the spread at entry, so its stop is effectively nearer and its target further away by the same amount.

Why it matters most for short-term strategies

Spread is a fixed cost per trade, but risk is not. The tighter the stop, the larger the share of it the spread consumes — which is why lower-timeframe and scalping strategies are the most exposed to an unrealistic cost model, and why they can look best in a backtest that ignores it:

A strategy that trades around the open, the close or scheduled releases is doubly exposed, because those are the hours when spread moves most. Measuring your results by hour and by session — which Edge Lab does on backtest sessions and on your live journal alike — is how you find out whether an attractive hour is still attractive after costs.

How Sage handles historical execution

• Bid candles, a stored spread per minute. Sage stores bid candles and, for each 1-minute candle, the ask price at that minute's close. The difference is that minute's spread, and the ask side of the candle is taken as the bid plus it.
• Fills inside the bar. Orders, stops and targets are judged on bars finer than the candle you are viewing — 1-minute bars when you replay on 1 minute, and never coarser than 5 minutes for any step up to 4 hours. Each of those bars carries the spread quoted at its close.
• Touch means filled. A level fills when its side of the book trades at it. If price gaps straight past a level between bars, it fills at the first price that existed after the gap, so a gapped stop can lose more than 1R, as it can live.
• Conservative when unsure. When one bar spans both the stop and the target, the stop is taken first.
• A fallback that is never free. Where no spread was stored for a minute, the session's fixed spread is charged and the workspace shows it — a minute with no data is not treated as a minute with no cost.
• Real spread is the default for a new session, and the session form shows which dates have spread history for the instrument. Crypto history in Sage has no ask side, so crypto sessions use a fixed spread.

What it does not model

Historical bid/ask data removes one assumption. It does not reproduce live trading, and it is worth being exact about what is still missing:

• Slippage beyond the spread. Fills happen at your level (or at the first price after a gap); there is no extra slippage on fast markets.
• Commissions, swaps and financing. Not charged in a replay. If your account pays commission per lot, results are optimistic by that amount.
• Latency, liquidity and market impact. Every order is filled in full at the quoted price, with no queue, no partial fills and no effect on the market.
• Spread inside a minute. The spread is recorded once per minute, at its close; a spike that came and went within a minute is not seen.
• Your broker's quotes. The spread is the data provider's, not your broker's. Brokers add markups or charge commission instead, so use a custom spread when you know yours.