Why I test gold EAs from 2022 – what short backtests don’t show you
Why I test gold EAs from 2022 – what short backtests don't show you
A beautiful curve is not proof
A gold EA that shows a perfect equity curve from 2024 onwards has proven very little. Since 2024 gold has been in one of the strongest bull markets in decades, and almost any long-biased system looks brilliant in that phase.
Over the past months I ran many gold EAs through the MetaTrader 5 Strategy Tester on real ticks, starting in January 2022 instead of 2024. The results changed my view of what a backtest can and cannot show. This post shares four lessons and a checklist you can use for any gold EA, including mine.
Why 2022 is the honest starting point
Since January 2022 gold has gone through three very different phases, and a robust EA has to survive all of them.
| Phase | Period | Character |
|---|---|---|
| Sideways | 2022 – early 2024 | choppy, many false breakouts, sharp swings in both directions |
| Strong trend | 2024 – 2025 | steady rise, shallow pullbacks |
| Volatile trend | 2026 | large moves, fast reversals |
A test that starts in 2024 skips the hardest phase completely. That is exactly the phase in which many systems lose money or break down.
Lesson 1: the same EA can look like a winner or a total loss
One grid EA I tested showed a profit factor above 2 and more than five times its starting capital from 2022 to 2026 on its lowest risk level. On its default risk level, the same EA lost the entire account.
The low-risk version only survived because its smaller lots gave it more margin to sit through the big adverse moves. In the strong trend years it earned everything; 2022 and 2024 were losing years. Another grid EA, published with a test starting in 2024, showed a profit factor of 5. Its mechanics were almost identical.
Takeaway: ask for the test from 2022. If it only starts in 2024, test it yourself – the Strategy Tester demo on MQL5 makes this free.
Lesson 2: look at equity drawdown, not balance drawdown
The balance only changes when a trade is closed. Equity also includes the open, floating losses. A system that keeps losing trades open shows a tiny balance drawdown and a huge equity drawdown.
In one report I analysed, the balance drawdown was 3 %, but the relative equity drawdown was 33 %. The account had been far deeper under water than the balance ever showed – the losses were simply not closed yet.
Takeaway: in the tester report, compare "Balance drawdown" with "Equity drawdown". A large gap is the signature of a system that sits out its losses.
Lesson 3: a high win rate can hide the real risk
Win rates of 75 % to 90 % look reassuring. In grid and averaging systems they usually mean one thing: losing positions are not closed but averaged down until the market comes back.
| What I measured in one grid EA | Value |
|---|---|
| Winning baskets | 798 |
| Losing baskets | 42 |
| Average winning basket | +41 USD |
| Average losing basket | −1,268 USD |
Three of those 42 baskets wiped out years of small gains. The report also showed a correlation of 0.90 between each trade's result and its maximum adverse excursion: the outcome depended almost entirely on how far the market ran against it.
Takeaway: look at average win versus average loss, the longest losing streak, and whether positions are added in the same direction at worse prices.
Lesson 4: compounding can hide losing years
A final balance of several hundred thousand dollars says nothing about how the account got there. One minute-scalper with a fixed stop loss showed a net profit of almost 500,000 USD from 20,000 USD. Year by year it looked very different:
| Year | Result |
|---|---|
| 2022 | −82 % |
| 2023 | −26 % |
| 2024 | −2 % |
| 2025 | +139 % |
| 2026 | +8,298 % |
The account fell to under 1,800 USD before a single exceptional year, with lot sizes growing to over 70 lots per trade, produced the entire profit. Results like these are neither repeatable nor executable in real markets.
Takeaway: always check the result per year and the deepest point of the account, not just the final number.
A checklist for any gold EA backtest
-
Test period: starts no later than January 2022 and includes the sideways phase.
-
Modelling: "Every tick based on real ticks". One-minute OHLC can make tight grids and scalpers look far better than they are.
-
Equity vs balance drawdown: a large gap means losses are being sat out.
-
Stop loss on every trade: check the trade list for positions without SL or for repeated entries in the same direction at worse prices.
-
Average win vs average loss: a 90 % win rate with tiny wins and rare huge losses is a warning sign.
-
Result per year: every year, not only the total. Note the longest period without a new equity high.
-
Lot size over time: compounding to very large lots inflates results that cannot be executed live.
-
Live track record: a public signal over at least 12 months beats any backtest.
How I applied this to my own EA
I built GBO – Gold Breakout Orchestra with exactly this checklist in mind: every trade has a stop loss, there is no grid and no averaging down, and every configuration was tested on real ticks from January 2022.
| GBO configuration | Max. equity drawdown | Profit factor | Positive years 2022–2026 |
|---|---|---|---|
| Trend Conservative (USD 1,000, fixed 0.01 lot) | under 12 % | 2.76 | 5 of 5 |
| Pump & Energy | 17.7 % | 2.21 | 5 of 5 |
| Punch & Weight | 36.9 % | 2.22 | 5 of 5 |
| Push Forward | 50.3 % | 1.98 | 5 of 5 |
The weak spots are part of the picture too. The conservative set went through 264 days without a new equity high from May 2023 to February 2024, and the gains in 2022 and 2023 were small: +8 % and +4 %. A trend-following system earns its money when gold trends, and it has to be patient when it does not.
Conclusion
The most important question about a gold EA is not how much it made in the best years, but how it behaved in the worst ones. Test from 2022, read the equity drawdown, check every year, and be suspicious of curves that look too smooth.
If you want to see how GBO handles these questions, the product page and the user manual are on my MQL5 profile. Test it yourself in the Strategy Tester from January 2022 on real ticks – that is the best way to form your own opinion.
Risk notice: trading leveraged products such as gold CFDs carries a high risk of losing capital. All backtest results are hypothetical, are based on historical data and do not guarantee future performance.
