Learn · Gold futures · Method
How to backtest gold futures manually (GC and MGC)
A practical method for testing a gold futures idea by hand on historical sessions — the same process works in any replay tool. The gold-specific parts are tick sizing, choosing clean sessions and handling rolls and data releases.
By the NQ Replay team · Updated
Short answer
Write your rules in ticks, list the sessions to test before you start, replay each one bar by bar without looking ahead, and log every signal in the same format. Exclude or tag holiday, roll and release sessions, then judge results in multiples of risk (R) — and treat a few months of gold data as a small sample.
The seven steps
1. Write the rule before you look
Entry trigger, stop, target and invalidation, all in ticks. Example form: "after X, enter on Y, stop Z ticks beyond the swing". If a rule needs judgement, write down how you'll judge it.
2. Choose the contract and convert to dollars
Rules in ticks stay the same; dollar risk differs tenfold between GC and MGC. Use the calculator to note the dollar risk per trade.
3. Build the session list first
List every session you'll test before replaying any of them, so you can't pick favourites. Decide up front how to treat holiday, roll and release days (below).
4. Replay bar by bar
Start each session at 6:00 p.m. ET or at your window, and step forward. Higher timeframes in NQ Replay are built only from bars up to the replay clock, so they don't leak the rest of the day.
5. Log every signal, including the ones you skip
Use a fixed journal template. Skipped signals tell you whether the rule is really mechanical.
6. Review in R, not dollars
Express each result as a multiple of the initial risk (R). Then look at count, win rate, average R, longest losing run and maximum drawdown in R.
7. Re-test on unseen sessions
Keep some sessions aside, or replay them later, to check the result wasn't tuned to the first batch.
Picking clean gold sessions (2026 archive)
The archive has available GC sessions from Jan 2, 2026 and MGC sessions from Feb 24, 2026, both to Aug 24, 2026 — 166 GC sessions, each holding one outright contract. Not every available session is a normal one. These are the ones to exclude or tag:
| Type | Sessions | Why |
|---|---|---|
| Holiday | Jan 19, Feb 16, Apr 3, May 25, Jun 19, Jul 3 | Shortened (or, for Good Friday, missing) sessions with unusual liquidity. |
| Partial | Feb 25 | No bars from about 1:00 to 2:45 p.m. ET. |
| Partial | Aug 24 | Only 6:00–8:00 p.m. ET on Aug 23 is stored. |
| Contract switch | Jan 28, Mar 27, May 27, Jul 29 | First session of a new contract; prior levels belong to the old contract. |
| Scheduled releases | CPI, jobs report, FOMC days | Tag them so you can compare results with and without event days. |
After those exclusions there are 159 full outright GC sessions between January 2 and August 21. Details on roll handling are in gold futures rollover; event days are covered in gold around CPI, jobs and FOMC.
Gold-specific mistakes
| Mistake | Fix |
|---|---|
| Copying NQ point distances | Size from gold's own range. The 2026 median GC session range was about $108/oz (1,081 ticks). |
| Ignoring the Asian and London hours | About 44% of median GC volume traded between 6:00 p.m. and 8:00 a.m. ET; decide whether your rules apply then. |
| Carrying levels across a roll | Redraw on the new contract or note the roll gap. |
| Assuming fills at the stop price | 1-minute bars can't show intrabar slippage; log a conservative fill when a bar spikes through your stop. |
| Treating months as years | A 2026-only gold sample is short. Report the number of trades with every result. |
What a manual backtest can and cannot show
It shows whether you can apply a rule consistently and how it behaved on a known set of past sessions. It cannot prove the rule will work in future, and simulated results leave out real-money pressure, queue position and slippage. Use it to refine a process, not to forecast returns.
Frequently asked questions
Can you backtest gold futures without coding?
Yes. Manual backtesting means replaying historical sessions bar by bar, applying written rules and logging each simulated trade. It is slower than code but makes you practise the decisions too.
How many trades do I need for a gold backtest?
There is no magic number, but a few dozen trades is a very small sample and results can change a lot with more data. With 159 full GC sessions in the 2026 archive, a once-a-day setup gives at most 159 trades.
Should I backtest on GC or MGC?
The chart is the same market, so use whichever matches how you would size. Write rules in ticks and convert to dollars: $10 per tick on GC, $1 on MGC. GC has more history in the archive (from January 2, 2026 vs February 24 for MGC).
What is lookahead bias?
Using information that would not have been available at the decision moment, such as the day's final high. In replay, keep future bars hidden and make every decision from the bars already printed.
Sources
- CME Group — Gold (GC) futures contract specifications
- CME Group — Micro Gold (MGC) futures contract specifications
- BLS — 2026 release schedule
- Federal Reserve — FOMC meeting calendar
- NQ Replay 1-minute archive (GC, MGC, NQ, MNQ), measured October 23, 2026. Measurements describe past sessions only.