QuantexMarket risk research

12 Oct 2026 · Study: Backtest versus live evidence · Case study 01

Seven months of history was not enough

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

Abstract

Can about seven months of history show whether a systematic gold model is robust? We rebuilt one of our own XAUUSD models as a simulation, checked it against its real history (it reproduced 69–77% of the largest real adverse events) and ran it with costs on 21 months and on 4.4 years of one-minute data, changing single rules to test sensitivity. The short sample overstated the result: almost all of the gain came from 2026, a year of unusually large moves, while 2025 was flat and 2023 negative in every variant. One small rule change turned the best variant negative and roughly doubled its worst drawdown. We withdrew an earlier risk setting that would have led to a drawdown of about 99%, and now calibrate risk on weaker variants with a margin of safety.

Keywords: backtesting; overfitting; sample length; market regimes; gold; XAUUSD
JEL: G11; G17; C52

1. Introduction

We re-tested one of our own systematic gold models on a much longer history. The short test we had relied on before made the model look stronger than it is. This note explains the method, the findings and what we changed.

The question

Can about seven months of history tell us whether a systematic model is robust?

2. Data and method

Method

  • We rebuilt one of our internal gold (XAUUSD) models as a simulation and checked it against its real history. On the largest adverse events, the simulation matched about 69–77% of the real ones.
  • We ran the same rules on two samples of one-minute market data: 21 months (January 2025 to October 2026) and 52.8 months (May 2022 to October 2026, about 1.56 million one-minute bars).
  • We included transaction costs, then changed single rules slightly to test sensitivity.

3. Results

Findings

  1. The short sample overstated the result. An earlier study on about seven months of data looked much stronger than the longer tests.
  2. The result came mostly from one regime. On 21 months, 2025 was flat for every variant tested. Almost all of the positive result came from 2026, a year of unusually large gold moves.
  3. Calm years behave differently. On 4.4 years, the best variant kept a small positive edge on average, but 2023 was negative in every variant we tried.
  4. The result was sensitive. One small rule change (the minimum distance between entries, set to $0.10) turned the best variant negative and roughly doubled its worst drawdown.

4. Discussion

What we changed

  • We withdrew an earlier risk setting. On this history it would have led to a drawdown of about 99%.
  • Risk is now set from a weaker variant of the test, not the best one, with an additional margin of safety.
  • We now treat calm years such as 2023 as the base case, and high-volatility years such as 2026 as the exception.

Lessons for researchers

  1. Test across several years and market regimes, not only the recent past.
  2. Change each rule slightly. If the result breaks, the effect is weaker than it looks.
  3. Calibrate risk on the weaker cases you found, and add a margin of safety.
  4. Then verify on new data and in live conditions, with a record others can check.

Next in this study

We will compare the simulation with the live record month by month, and publish where they agree and where they do not.

5. Limitations

Limits of this study

A simulation is a model, not a record. Our simulation did not match every real event, and real-world execution can differ. Past results, real or simulated, do not guarantee future results.

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, 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). Seven months of history was not enough. Quantex Research, AlgoTrade Pte. Ltd. https://quantex.com.sg/research/2026-10-12-history-vs-live/

All notes and case studies

For education and information only. Not investment advice. See our Disclaimer.