Gold RSI Overbought Signal: 2-Week and 6-Week Forward Returns Analysis

1. Data & Confidence Context

The signal engine analyzed 100 confirmed RSI-crosses-above-70 events in GC=F spanning January 2001 through June 2026 — a meaningful sample by technical backtest standards, though gold's behavior across this period was shaped by at least three structurally distinct macro regimes (the post-dot-com dollar decline, the post-GFC zero-rate era, and the post-2022 inflation/geopolitical cycle). Forward-return statistics are engine-computed from actual crossing dates, not training-memory estimates, with 14-day and 45-day windows both carrying p-values well above conventional significance thresholds (p=0.36 and p=0.12, respectively), meaning the excess returns above the unconditional base rate cannot be statistically distinguished from noise at standard confidence levels. The honest framing: 100 crossings is a workable sample, but the signal's edge — if any — is modest and inconsistent enough that position sizing should reflect genuine uncertainty rather than a presumed directional edge.

---

2. Direct Answer — What the Data Shows

The engine's verdict is clear and worth sitting with before building any narrative around it: across 100 RSI-above-70 crossings in GC=F from 2001 to June 2026, the 14-day mean forward return was +0.31% with a median of +0.75% and a win rate of 57.0%. The 45-day mean was +0.90% with a median of +0.74% and a win rate of 54.0%. Against unconditional base rates of +0.42% (14-day) and +1.46% (45-day), the signal actually *underperforms* the base rate by -0.11% at two weeks and -0.56% at six weeks. Neither excess return clears statistical significance.

What does that mean in practice? When gold's RSI crosses above 70, you are not entering a reliably bullish setup — you are entering a moment where gold has already moved hard enough to register overbought, and the subsequent 45 days look almost indistinguishable from any random 45-day window in gold's history, except slightly worse on a risk-adjusted basis.

The chronological story of how this plays out is instructive. The earliest crossings in the dataset — August and September 2001, then the cluster running from January through May 2002 — came as gold was awakening from a two-decade bear market. The August 17, 2001 crossing preceded the September 11 attacks by less than a month; the September 14, 2001 crossing landed three days after them. In those instances, the RSI signal was essentially a coincident indicator of a fear-driven spike, and the subsequent 14-day windows were volatile and directionally mixed as the initial shock faded. The December 2002 and January 2003 crossings arrived as the dollar began a sustained decline and the Iraq War drumbeat grew louder — here, the signal coincided with a genuine trend, and forward returns were positive. But the signal didn't *cause* the trend; it merely flagged that the trend was already underway.

Fast-forward to the 2008-2011 period, when gold's secular bull market generated some of the most dramatic RSI crossings in the dataset. Crossings during this era often preceded further gains — but they also preceded sharp mean-reversion episodes, particularly in 2008 when gold spiked on Lehman-era fear and then sold off hard as dollar liquidity demand overwhelmed the safe-haven bid. The best single-instance 14-day return in the dataset was +8.96%; the worst was -6.77%. At 45 days, the range widens to +16.52% on the upside and -15.18% on the downside. That spread — nearly 25 percentage points at the 45-day horizon — is the defining feature of this signal. It is not a consistent edge; it is a wide distribution with a slightly positive center of mass that sits *below* what gold delivers unconditionally.

The win rate of 57% at 14 days sounds encouraging until you account for the magnitude asymmetry: the losing instances tend to be sharper and faster than the winning ones, which is why the mean (+0.31%) lags the median (+0.75%) and why excess return is negative. The 45-day picture is similar — 54% win rate, but the mean is dragged down by the left tail.

---

3. Confounding Factors — Decomposing What Actually Drove Each Cycle

The reason this signal's edge is so thin is that RSI crossing 70 in gold is not a single phenomenon — it is a label applied to at least four structurally different situations, each with different forward dynamics.

The first is the fear spike: gold surges on a geopolitical or financial shock (September 2001, Lehman 2008, COVID March 2020, the 2022 Ukraine invasion onset). In these cases, the RSI crossing is a coincident indicator of peak fear, and the subsequent 14-45 days often see mean reversion as the acute shock fades and dollar liquidity demand reasserts itself. The dollar's behavior is the key sequencing variable here — in the first two to four weeks after a fear spike, if the dollar strengthens (as it did sharply in late 2008), gold's RSI-overbought condition resolves to the downside. The worst instances in the dataset cluster around this dynamic.

The second is the trend acceleration: gold is already in a multi-month uptrend, RSI crosses 70 as momentum builds, and the crossing marks not a top but a mid-trend continuation. The 2002-2003 crossings and the 2010-2011 crossings fit this pattern. Here, the confounding force is dollar weakness — when the trade-weighted dollar is in a sustained downtrend, RSI-overbought conditions in gold tend to resolve sideways-to-higher rather than sharply lower, because the fundamental driver (dollar depreciation) is still intact. In these instances, the 45-day return was meaningfully positive, but the signal was doing little work — the dollar trend was doing the heavy lifting.

The third is the inflation-fear regime, most visible in 2022-2023 crossings, where real rates and CPI expectations were the dominant force. When real rates were deeply negative, RSI crossings in gold tended to resolve higher; as real rates rose sharply through 2022-2023, crossings that would have been bullish in a prior regime became exhaustion signals. The sequencing mattered enormously: in the first three months of a rate-hiking cycle, gold's RSI crossings were more likely to mark short-term tops; in months six through twelve, as rate-hike expectations became priced in and the dollar peaked, the same signal became more reliably bullish.

The fourth is the low-volatility grind, where gold drifts higher on modest volume, RSI crosses 70 without a dramatic catalyst, and the subsequent 45 days are essentially random. These are the crossings that most dilute the signal's aggregate statistics.

---

4. What This Means Now — Scenario Analysis

As of June 2026, the chart covering January 2001 through June 26, 2026 shows the full arc of gold's history across these 100 crossings. The current moment's most relevant historical analog depends on which macro regime is dominant — and that is precisely the variable the RSI signal cannot tell you.

Scenario A — Fear-spike analog (worst-case for the signal): If the most recent RSI crossing above 70 was driven by a geopolitical or financial shock rather than a sustained trend, the historical base case is a 14-day return near flat to slightly negative, with meaningful downside risk (worst case in the dataset: -6.77% at 14 days). The key tell would be dollar behavior in the first week post-crossing: a strengthening dollar in days 1-7 has historically been the leading indicator of the left-tail outcome.

Scenario B — Trend-continuation analog (best-case for the signal): If gold is in a multi-month uptrend with a weakening dollar and negative or declining real rates, the RSI crossing is more likely to resolve in the upper half of the distribution. The best 45-day outcome in the dataset was +16.52%, achieved in trend-continuation regimes. The key tell here is whether the dollar index is making lower highs and whether real rates are flat-to-declining.

Scenario C — Base case (most probable given aggregate statistics): The 14-day return lands somewhere between -2% and +3%, consistent with the mean of +0.31% and the wide distribution. The signal provides no reliable directional edge, and the 45-day outcome is similarly indeterminate, with the unconditional base rate of +1.46% actually *exceeding* the signal's mean of +0.90%.

The variable that determines which scenario plays out is not the RSI level itself — it is the dollar trend and real rate direction in the two weeks following the crossing.

---

5. Actionable Implications — With Explicit Uncertainty

Given the p-values of 0.36 (14-day) and 0.12 (45-day), neither return series clears statistical significance. Any position taken on this signal alone should be sized as a *speculative* trade, not a high-conviction systematic bet.

Causal mechanism identified: RSI crossing above 70 in GC=F signals that gold has experienced a sharp near-term move. It does not reliably predict continuation or reversal — it predicts elevated volatility relative to the base rate, with a wide return distribution (best: +8.96% at 14d, worst: -6.77% at 14d).

Conditions under which the bullish case holds: Dollar in a downtrend, real rates flat or declining, crossing occurs within an established multi-month uptrend. Under these conditions, historical analogs suggest the upper half of the distribution is more accessible.

Conditions under which it breaks down: Dollar strengthening post-crossing, real rates rising, crossing driven by a fear spike rather than trend momentum. Under these conditions, the left tail (-6.77% at 14d, -15.18% at 45d) becomes the operative risk.

What to watch in the first 5 trading days: Dollar index direction is the single most important confirming or disconfirming variable. A dollar that strengthens in the first week post-crossing has historically been the clearest leading indicator of a negative 14-day outcome for gold.

Position sizing implication: Given a 57% win rate at 14 days and negative excess return, this signal does not justify outsized position sizing. If used at all, treat it as a secondary confirming signal within a broader framework — not a standalone entry trigger. Risk no more than you would on a coin-flip trade with a slight positive skew, because that is approximately what the data supports.