We backtested 9 commodity technical signals against 25+ years of data. Most failed to reach statistical significance. Here is what the data actually shows, and what it means for traders.
2026-06-18 · 8 min read
We backtested 9 commodity technical signals against 25+ years of data. Five passed. Four failed. The failures aren't obscure signals — they include the gold golden cross, gold RSI overbought, and the copper death cross. Signals that commodity traders discuss constantly, and trade on regularly.
That's the uncomfortable finding: most of the signals traders treat as meaningful aren't. The patterns are real. The predictive power isn't.
The gold golden cross is the best example of how traders fool themselves. Over 25+ years, it fired 15 times. At 180 days, 73.3% of those events resolved higher. That's the highest win rate in our entire dataset — and it's also not statistically significant.
The p-value is 0.21. That means there's roughly a 1-in-5 chance of seeing a 73% win rate across 15 events even if the signal has zero predictive power. It's not that the result is wrong — it's that the sample is too small to tell signal from luck. With 15 events in 25 years, you simply don't have enough data.
This is the trap: a high win rate over a small sample is not evidence. It's an invitation to investigate further. A 73% win rate over 15 events looks compelling. A 60% win rate over 80 events might be far more significant. The percentage alone doesn't tell you which one to believe — sample size does.
P-value is the probability that your backtest result occurred by chance. A p-value of 0.05 means a 5% chance the result is coincidental. The conventional scientific threshold for “significant” is p < 0.05.
For traders: if you flip a coin 15 times and get 11 heads, that's a 73% heads rate — but it's not surprising, it happens regularly with fair coins. If you flip 1,000 times and get 73% heads, you start to believe something real is going on. P-value captures the difference mathematically.
The practical rule: never evaluate a backtested signal without knowing how many times it fired over the test period. A headline win rate without a sample size is nearly useless.
Of the five signals that reached significance, several are counterintuitive enough that they're worth examining closely.
Gold RSI oversold (p=0.0013) is the strongest signal in our dataset. When gold's RSI drops below 30, price is higher 71.8% of the time 30 days later, across 39 historical events. The conventional read of an RSI oversold reading is “more downside possible.” The data says buy.
Gold death cross (p=0.03) is stranger. The “bearish” crossover — 50-day MA falling below 200-day MA — predicts higher gold prices 75% of the time at 180 days. Not lower. Higher. The most plausible explanation: gold has a persistent long-term uptrend, so death crosses tend to mark temporary oversold corrections rather than genuine reversals. The signal that every chart reader calls bearish turns out to be one of the more reliable buy signals we've found.
Crude oil death cross + RSI below 35 (p=0.003) is the most robust combined signal in the dataset. When crude forms a death cross while simultaneously oversold, price resolves higher 65% of the time at 180 days across 46 events. The combined condition filters false positives that either trigger alone would produce.
Gold RSI overbought: 51% win rate at 90 days, p-value of 0.29–0.52. The median 90-day return is slightly negative. As close to a pure coin flip as any signal we've tested. Overbought RSI in gold adds no information.
Copper death cross: 57% win rate at 180 days, p-value of 0.36–0.59. The win rate is actually below the unconditional base rate for copper over that period — meaning you would have done slightly better just buying and holding without any signal.
These aren't rare or obscure signals. They're actively discussed in commodity analysis, appear in technical alerts, and generate real trading decisions. The data says they don't predict anything.
From our testing: at minimum 20 events over the test period, a p-value below 0.05, and consistency across multiple time horizons — not just the one window that looks best. Natural gas RSI oversold is significant at 30 days (p=0.016) but the edge fades by 90 days. That's still a useful finding: trade it short-term, not as a long-term hold. But it's also a warning about horizon-shopping — if a signal only works at one specific horizon, that's a red flag.
The other useful filter: does the result make sense mechanically? The gold death cross being contrarian bullish is counterintuitive, but it has a plausible explanation given gold's long-term trend structure. The copper death cross having no edge is also plausible — copper is more macro-driven and mean-reverting than gold. When the data and the mechanism align, confidence increases. When a signal only works in one arbitrary window with no coherent explanation, be skeptical.
The nine signals we've published are a starting point. Every commodity has dozens of potential signals worth testing. The ones that pass are valuable precisely because so many fail — signal scarcity is what creates edge.
You can check any of our tested signals instantly at seeerai.com/check. For signals we haven't tested yet, the free 7-day trial includes 10 analyses — ask anything in plain English. “Does a silver death cross predict lower prices at 90 days?” is a legitimate query, and it'll return a p-value, win rate, and a verdict you can actually act on.
The most expensive trading mistake is building a position around a signal that looks compelling but has no statistical basis. Testing it costs less than one bad trade.
Test any commodity signal
Ask any question about commodity signals or macro events — Fed decisions, ECB rate hikes, supply shocks, tariff cycles — in plain English. Statistical significance testing included.
Start Free Trial →