The overnight gold breakout made almost all its money in a single quarter, when gold was moving ten times more than a year earlier; without that quarter, it loses. And of 88,128 technical systems tested on nine markets, not one passed on seven of them. Something that only works on one market (or for a few months) should make you suspicious.
17 technical signals on 9 markets, with a chance control
Moving average crosses, Donchian breakouts, Bollinger, RSI, MACD, SuperTrend, Keltner and VWAP: 17 classic signals, with and against the move, with different stops, targets and time windows. 2,448 variants per market and timeframe, 88,128 systems in total, all closed at the end of the session and with costs. For each one, a twin with the same hours but a randomly chosen direction.
On seven of the nine markets, not a single variant passed, not even the random ones: with costs, they lose or are indistinguishable from zero. On the Nasdaq 64 passed versus 11 from chance, and all of them were trend following. Only 1 of the 3 families examined then passed on years nobody had looked at.
With another engine, 59,400 simple one-day rules produced 269 candidates on the Nasdaq versus 25 from chance. On gold, the 10-year note and the euro, only between 0.0% and 0.7% of the rules made money: below chance.
The Nasdaq strategies, on the S&P
We took three strategies that do hold up on the Nasdaq to the S&P 500 without touching a single parameter:
| 1 contract | Micro S&P 2019-2026 | S&P 2006-2019 (unseen years) |
|---|---|---|
| Nasdaq strategy A | −$3,126 | −$1,421 |
| Nasdaq strategy B | +$6,845 | −$7,664 |
| Nasdaq strategy C | −$1,718 | −$3,113 |
None of them transfers. The only one that made something on the recent S&P lost clearly in the older years.
Gold: it's not the costs
We took the three families that had done best on the Nasdaq to gold (about 907,000 1-minute bars from 2024-2026): 92 settings in four time windows. None transferred. The decisive test was repeating it with no costs: the gross profit factor was between 1.03-1.14 in the COMEX session and 0.95-0.98 in the London session. In London it loses even without paying commissions: that edge doesn't exist on gold, and switching to the micro (MGC) doesn't fix it.
Overnight gold: a bet on volatility
From 7 to 9 PM New York time, breakouts did show something: a gross profit factor of 1.14-1.18, which after costs stayed at 1.03-1.06. With a target equal to the stop you needed a 52.7% win rate and got 48.2%.
The important part came from looking quarter by quarter: the correlation between the quarter's volatility and profit was 0.76.
- In 2024, with gold moving 0.35-0.43 pts/min on average, it lost all four quarters.
- In the first quarter of 2026, with 4.35 pts/min, it made +$4,453: practically all the profit of the period.
- Without that quarter, the system loses.
That's not a strategy: it's a bet that gold stays very volatile. If volatility goes back to 2024 levels, expect to lose again.
A trap we fell into: filtering "wide opening range" days seemed to improve results a lot. It was a mirage: the range size was measured with the distance between each trade's entry and exit, something you only know after closing it. With a measure that is known at entry, the effect disappeared.
Why "it only works on the Nasdaq" is suspicious
There are two possible explanations: a mechanism specific to the Nasdaq (a dominant cash session, fund flows concentrated in market hours) or a trait of some very bullish and volatile years. Looking only at recent Nasdaq you can't tell them apart. That's why we demand other years of the same market, a chance control and similar markets.
- If it works on a single market, require it to work on other years of that market, with a crisis or a sideways period inside.
- Test it with no costs: if it loses even then, there's nothing to optimize.
- Look at your results quarter by quarter next to volatility. If all the profit is in one or two extreme quarters, you have a bet, not a strategy.
- Don't switch markets hoping your strategy will work there: it usually works worse.
- Use in your filters only information you knew at entry.
How it was calculated
Systems closed at the end of their session, with a $0.62 commission per side and 1 tick of slippage per side, fitting, validation and test periods, and a random-direction twin for each system. The three Nasdaq strategies were moved to the S&P without changing parameters, decided before looking. Lab results, simulated and hypothetical.
Simulated and hypothetical results, with the limitations that implies. Not investment advice. Firm rules change: check them on each firm's website. This article contains no affiliate links.