QuantexMarket risk research

17 Oct 2026 · Updated 21 Oct 2026 · Study: Backtest versus live evidence · Research note

What the 30 January 2026 gold drop did to EAs

Huyen X. Vu · Quantex Research, AlgoTrade Pte. Ltd., Singapore · ORCID 0000-0002-2209-842X

Abstract

On 30 January 2026 spot gold fell about 9.5% to 9.8%, reported as its largest one-day fall since 1983, a day after a record of about US$5,595 an ounce; CME then raised gold futures margins from 6% to 8%. We set out the publicly reported facts and use the day as a test case for EA design: grid and averaging EAs, martingale and lot-multiplying EAs, and stop-loss EAs face different risks on a one-way day of this size. We list what a risk check should include so that the next such day is not a surprise, and state what public sources do not report.

Keywords: gold; XAUUSD; market crash; 30 January 2026; grid trading; martingale; margin
JEL: G01; G11; Q02

1. Introduction

On Friday 30 January 2026 gold had its largest one-day fall in more than four decades. This note sets out what is publicly reported about the day, then looks at it as a test case: what different kinds of Expert Advisor (EA) face on a day like this, and how to check a model against it.

The question

When gold falls this far in one session, which EA designs are most exposed, and what should a risk check include so that the next day like this is not a surprise?

2. Results

What happened: the reported facts

Not found in public sources: the exact spot low and close for the day, the hour the selling started, and any broker-level changes to CFD margin or spreads. We do not estimate them here.

3. Discussion

What different EA designs face on a day like this

Grid and averaging EAs

A grid adds positions as price moves against it and waits for a pullback. A one-way move of this size fills level after level in one session, so the open loss and the margin used grow together. The model's survival depends on the distance between levels, the lot steps and the free margin at the start of the day, not on its average past result.

Martingale and lot-multiplying EAs

When lots grow after each loss, a long run of adverse levels grows the exposure faster than the distance. This is the design most likely to meet a margin call on a day of this kind.

Stop-loss EAs

A stop limits the loss only if it fills near its price. In fast markets, fills can be worse than the stop price, and spreads can widen. A model tested with a fixed spread and perfect fills will look safer than it was.

Trend-following EAs

A short-side trend model can gain on the day itself. The risk moves to what comes next: after a record move, sharp reversals and wide ranges can produce a run of false signals.

Margin changes after the event

Exchange margin rose after the fall. Brokers set their own CFD margin, and we found no published broker changes for these days. Any EA that runs close to its margin limit should be checked against a higher margin requirement, not only against price.

How far can a grid survive? A worked example

A $10,000 account runs a buy grid: 0.01 lot at the first level, one more position at every fixed step as price falls, and no take-profit on the way down. Spread, swap and the broker's stop-out are left out, so real results are worse.

How far gold must fall to wipe a $10,000 grid accountGold fall that wipes a $10,000 grid account: $10 step flat lot $442, ×1.2 $199, ×1.5 $126, ×2.0 $90; $20 step flat lot $622, ×1.2 $330, ×1.5 $221, ×2.0 $160. The 30 January 2026 fall was about $520 to $540.How far gold must fall to wipe a $10,000 grid accountBuy grid from 0.01 lot, one position per step, no take-profit on the way down. Spread, swap and stop-out ignored.010020030040050060070030 Jan 2026 fall ≈ $520–540$10 step · flat lot$442 · 8.0%$10 step · ×1.2 lot$199 · 3.6%$10 step · ×1.5 lot$126 · 2.3%$10 step · ×2.0 lot$90 · 1.6%$20 step · flat lot$622 · 11.3%$20 step · ×1.2 lot$330 · 6.0%$20 step · ×1.5 lot$221 · 4.0%$20 step · ×2.0 lot$160 · 2.9%Gold price fall (USD) before the floating loss reaches $10,000Arithmetic illustration on a straight-line fall, not a test of any EA. % of a $5,500 reference price. Fall size as reported by Reuters.Quantex ResearchHow far gold must fall to wipe a $10,000 grid accountGold fall that wipes a $10,000 grid account: $10 step flat lot $442, ×1.2 $199, ×1.5 $126, ×2.0 $90; $20 step flat lot $622, ×1.2 $330, ×1.5 $221, ×2.0 $160. The 30 January 2026 fall was about $520 to $540.How far gold must fall to wipe a$10,000 grid accountBuy grid from 0.01 lot, one position per step, notake-profit on the way down. Spread, swap and stop-outignored.010020030040050060070080030 Jan 2026 fall ≈ $520–540$10 step · flat lot$442 · 8.0%$10 step · ×1.2 lot$199 · 3.6%$10 step · ×1.5 lot$126 · 2.3%$10 step · ×2.0 lot$90 · 1.6%$20 step · flat lot$622 · 11.3%$20 step · ×1.2 lot$330 · 6.0%$20 step · ×1.5 lot$221 · 4.0%$20 step · ×2.0 lot$160 · 2.9%Gold price fall (USD) before the floating loss reaches $10,000Arithmetic illustration on a straight-line fall, not a test of any EA.% of a $5,500 reference price. Fall size as reported by Reuters.Quantex Research
  • A flat grid with a $10 step is gone before the day's fall is complete.
  • A flat grid with a $20 step survives on paper, with little left once a real stop-out applies.
  • Any lot multiplier wipes the account after a 2% to 6% move, a fraction of that one session.

Survival is set by three numbers chosen before the day starts: step size, lot growth and free margin.

Our own evidence

We did not re-run our models on this day for this note, so we give no new numbers. What our Case study 01 already showed:

  • Almost all of one systematic gold model's positive result on 21 months came from 2026, a year of unusually large gold moves. A model that earns from large moves is also exposed to them.
  • An earlier risk setting was withdrawn because, on the long history, it would have led to a drawdown of about 99%. Risk is now set from a weaker variant with a margin of safety.

Checklist: test your EA against this day

  1. Include the day. Make sure the test data covers 29 January to 2 February 2026, on real ticks if your broker provides them.
  2. Use today's size. Run the test at the lot size and account balance you use now, not those of the original test.
  3. Add worse fills. Widen the spread and add slippage for the session of the fall.
  4. Raise margin. Re-run with a higher margin requirement and check the free margin at the worst point.
  5. Check silver and other metals separately. The silver move was much larger in percentage terms.
  6. Read the method. Our note How to stress-test a gold EA for crash scenarios sets out the full method.

4. Limitations

Limits of this note

The event facts come from the public sources linked above; figures differ slightly between sources, so we give ranges. The EA section is a method, not a test of any one EA, and the grid example is arithmetic on a straight-line fall.

Appendix

FAQ

How big was the gold drop on 30 January 2026?
Spot gold fell about 9.5% to 9.8% in one day, after a record high of about $5,595 an ounce the day before. Reuters reported it as the largest one-day percentage fall since 1983.

Why did gold fall on 30 January 2026?
Reports point to the nomination of Kevin Warsh as Fed chair, a stronger dollar, and profit-taking after a crowded rally. CME margin increases announced in the following days were cited as adding to the selling in the days after.

Which EAs were most at risk?
Designs that add positions or grow lots as price moves against them, such as grid and martingale EAs, and any EA running close to its margin limit.

Related notes

Declarations

  • Author: Huyen X. Vu, Quantex Research, AlgoTrade Pte. Ltd., Singapore (ORCID 0000-0002-2209-842X).
  • Data availability: Public sources are listed under Sources. Results for our own models come from internal records that are not public; summary figures are shown in the note.
  • Use of AI tools: Data processing, analysis code and drafting were assisted by AI tools (Claude, Anthropic). The author checked the results and is responsible for the content.
  • Conflict of interest: The author is the founder of AlgoTrade Pte. Ltd., which develops and runs systematic trading models.
  • Funding: No external funding.
  • Peer review: Not yet peer reviewed. Comments are welcome at info@quantex.com.sg.
  • How to cite: Vu, H. X. (2026). What the 30 January 2026 gold drop did to EAs. Quantex Research, AlgoTrade Pte. Ltd. https://quantex.com.sg/research/2026-10-17-gold-30-jan-2026-crash/

All notes and case studies