Market risk researchHow deep can a gold grid basket go? Eight years of baskets, and the live record
Huyen X. Vu · Quantex Research, AlgoTrade Pte. Ltd., Singapore · ORCID 0000-0002-2209-842X · Download PDF
Abstract
The risk of a grid model sits in its few deepest baskets, which a short record can hide. We replayed the model's order rules on one-minute gold (XAUUSD) prices from 2018 to October 2026 (about 155,000 baskets) and measured depth three ways: number of orders, worst move against the first order, and time open. Since 2025 most baskets close within minutes and the 99th percentile is 21 orders, against 40 to 44 before, but moves against are larger in dollars: the widest move, US$406 per ounce on 29 January 2026, took only 22 orders in 87 minutes, while the basket with the most orders (342, October 2023) stayed open almost 29 days. The live tests ask whether no more than 2% of live baskets go beyond the replay's 99th percentile in orders or time.
Keywords: grid trading; gold; XAUUSD; tail risk; drawdown; replay; live test
JEL: G11; G17; G32
1. Introduction
The risk of a grid model sits in its few deepest baskets. A short record can hide them. We replayed the model over eight years of gold prices to see how many orders, how many dollars against and how many days a basket can take, and we check the live account against those numbers.
Where this project stands
| Stage | What it means here | Status |
|---|---|---|
| 01 Review | Question and data | Done |
| 02 Backtest | Replay of the model on one-minute prices, January 2018 to 9 October 2026 | Done |
| 03 Fine-tune | Not used: the model itself is not changed in this study | Skipped |
| 04 Live test | Live watch of the running account, with tests set in advance. Watch only: the account trades as before. | Now, from 23 Oct 2026 |
| 05 Handed over | Only if a finding is confirmed and a later study shows it is worth using | Not started |
Research question
How large do baskets get, measured three ways: number of orders, the worst move of gold against the first order (in US dollars per ounce), and time from first order to close? Does the live account stay inside the replay's range?
2. Data and method
The model we watch
Altair, a gold (XAUUSD) grid model shown on the AlgoTrade live board and studied in 26.07.ALW. When price moves against it, it adds orders on the same side, each larger than the one before, and closes them together. One such group is a basket. As with all our models, we do not publish its rules.
Data and replay
The backtest replays the model's order rules, as seen in its live orders, on one-minute gold prices: January 2018 to December 2021 from one price source (fixed spread) and May 2022 to 9 October 2026 from a broker's own one-minute history with its recorded spread (January to mid-May 2022 is not covered). Orders over the broker's 500-lot limit are capped at 500 lots. The replay was checked in two steps. First against a tick-by-tick platform run of the same rules on the same broker: for 2025 both had the same deepest basket (105 orders) and a net result within about 7% of each other. Then the platform run against the live account for January to 9 October 2026: 53,020 baskets against 47,072, and a deepest basket of 92 orders against 84. Most of that gap is from before 12 August 2026, when the model's order-size setting changed: before that date the live account had 87% of the platform run's baskets, after it 96%. So most of the gap comes from the replay using today's setting; the rest from price feed and fill differences. Swap and the live account's exact fills are not modelled. The model's order-size setting changed on 12 August 2026; the replay uses the current setting for the whole period. Price extremes inside a minute come from one-minute highs and lows.
Periods: results are shown for 2018–2021, 2022–2024 and 2025–October 2026 separately. A pattern counts only if it holds in all three.
Live counting. The live report counts only baskets whose first order is on or after 23 October 2026. A basket still open counts as soon as it has passed a test's limit; other open baskets wait until they close. The final read covers baskets that start in the first 15 weeks and is taken one week later, so that almost all of them have closed; any still open are listed. Days and times follow the broker's server clock (UTC+2 in winter, UTC+3 in summer).
How the live tests are read. The main 95% ranges treat baskets as independent. Baskets bunch together in busy weeks, so the final report also gives each range from resampling whole weeks (Monday to Sunday, broker time; 1,000 resamples with replacement; the middle 95% of the results), and a test counts as confirmed only if it passes on both. A basket counts as closed if it closed before the cut-off time of the final read; a position is one order when it opens, however it is filled. The test plan on this page was fixed on 23 October 2026, before any live basket, and a dated copy is kept in our research register. Results will be published whether they confirm, fail or stay unclear.
What we test live (set 23 Oct 2026, before any live case)
| Test | Backtest, 2025–26 | Confirmed if |
|---|---|---|
| H1: the live tail is within our tolerance of twice the replay's: share with more than 21 orders (the replay's 99th percentile) | 0.93% (831 of 89,797) | live share at most 2% and upper end of its 95% range at most 2% |
| H2: the same for time: share open longer than 1.6 hours | 0.94% (847 of 89,797) | live share at most 2% and upper end of its 95% range at most 2% |
Sample and timing. About 975 baskets a week in 2025–26 and about 1,180 a week in the live account in 2026, so roughly 15,000 after 15 weeks. If the true share is about 1%, as in the replay, the upper end of the basket-level 95% range would be near 1.2%, treating baskets as independent. Drawing 15 whole weeks at random from the 93 weeks of 2025–26 replay (2,000 draws) and applying both the basket-level and the week-level ranges, H1 passes in about 96% of draws and H2 in about 91%. Each share is counted over closed baskets plus open baskets that have already passed the limit. Open baskets that have not passed it at the final read are listed, and the report also gives the share if all of them were counted as past the limit, as an upper bound. A pass supports only these two tested rates, not the stability of basket risk as a whole. A pass means the tail frequency stays within this tolerance; it does not mean the live distribution matches the replay, and it says nothing about money. The move against in dollars is reported live but is not one of the tests. Live worst moves come from one-minute prices recorded on the account. Final read in early February 2027; the deepest live baskets are listed in the live report above as they happen.
3. Results
What the backtest found
| 2018–2021 | 2022–2024 | 2025–Oct 2026 | |
|---|---|---|---|
| Baskets (per week) | 34,502 (166) | 30,855 (225) | 89,797 (975) |
| Orders | 3 / 16 / 44 / 177 | 3 / 15 / 40 / 342 | 1 / 6 / 21 / 105 |
| Worst move against, US$ per ounce | 1.2 / 5.4 / 16.3 / 120 | 1.2 / 5.3 / 16.8 / 141 | 1.5 / 7.5 / 26.9 / 406 |
| Hours open | 0.15 / 1.7 / 11.7 / 187 | 0.13 / 1.2 / 6.9 / 690 | 0.02 / 0.22 / 1.6 / 73 |
| First order | Side | Orders | Worst move against | Open for |
|---|---|---|---|---|
| 12 Oct 2023 | Sell | 342 | $141 | 28.8 days |
| 14 May 2018 | Buy | 177 | $39 | 7.8 days |
| 11 Jun 2021 | Buy | 140 | $55 | 3.2 days |
| 15 May 2025 | Sell | 105 | $98 | 10.0 hours |
| 19 Mar 2026 | Buy | 73 | $333 | 7.7 hours |
| 29 Jan 2026 | Buy | 22 | $406 | 1.5 hours |
1. Depth in orders and depth in dollars are different things. The basket with the most orders (October 2023) moved $141 against over four weeks of slow, steady drift. The largest dollar move (January 2026) took only 22 orders in 87 minutes, because the price fell faster than one order a minute could follow.
2. Fewer orders is not lower risk. Since 2025 most baskets close within minutes and the 99th percentile is 21 orders, against 40–44 before. But moves against are larger in dollars per ounce, and a single slow trend, as in 2023, can keep one basket open for weeks. Part of the change may come from the replay using today's order-size setting for every year. The largest basket in each period is one observation, not a limit.
Live report
Rebuilt from the running account's orders, recorded to the second. Tables refresh every 5 minutes.
Loading the live report…
Interim reports
Week 1, data to 23 Oct 2026, 20:56 UTC. One trading day (23 Oct), 223 closed baskets. This is an early read, not a result: the first look is on 18 Nov 2026 and the final read around early February 2027. The live tables above keep updating; the numbers below are frozen at the cut-off.
Tail share. 1 of 223 baskets (0.45%) went beyond the replay's 99th percentile for the number of orders; the upper bound is 2.5%, above the 2% limit set in advance. Too early to say. The deepest basket had 27 orders and stayed open 1.25 hours (above the replay's 21-order 99th percentile, below its 1.6-hour mark). The largest worst move was US$36.31 (a 21-order basket open 1.11 hours), well below the US$100 flag.
4. Limitations
Scope: what this study cannot tell
- It describes the past. A deeper basket than any in this record is possible.
- It measures baskets, not account drawdown or money: the same move against costs very different amounts depending on order sizes, account size and margin.
- The 2018–2021 prices come from a different source with a fixed spread.
- The replay uses today's order-size setting for all years. Under the same setting, from 12 August 2026, it makes about 4% more baskets than the live account. The live test, not the replay, is the check.
Declarations
- Author: Huyen X. Vu, Quantex Research, AlgoTrade Pte. Ltd., Singapore (ORCID 0000-0002-2209-842X).
- Data availability: Live results are published in aggregate on this page and updated every five minutes. Order-level data can be shared with reviewers under a confidentiality agreement. Model rules are not published.
- 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, including the model studied here.
- 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). How deep can a gold grid basket go? Eight years of baskets, and the live record. Quantex Research, AlgoTrade Pte. Ltd. https://quantex.com.sg/research/altair-basket-depth/