The Gold What Every Score Meant Next: 55 Years of Point-in-Time Evidence
The Barometer's formula was reconstructed monthly from January 1971 to July 2026 using only data that would have been available at each date. For every observation the record shows what gold did over the following 1, 3, 5, and 10 years, by score band. Numbers on this page come from that reconstruction; the failures below are quantified rather than glossed.
Headline findings (read this first)
- The top zone (80 to 100 "Historically very favorable") has N=0 across the full 1971-present window. The composite never crossed 80 in 667 monthly observations. That zone is empirically UNTESTED and the site does not publish forward statistics for it.
- Band ordering is inverted at the 1-year horizon. Median nominal 1-year returns: Favorable 5.2%, Mixed 4.6%, Unfavorable 8.6%. Short-run gold moves and the score point in opposite directions on median. Read as a tactical 1-year buy signal, the instrument fails.
- The long-horizon real thesis does hold. Median 5-year real returns: Favorable 61.8%, Unfavorable -13.9%. Buyers who entered when the instrument read Favorable historically preserved and grew purchasing power over five years; buyers who entered when it read Unfavorable historically lost real ground.
- Real-rates sub-score correlation to subsequent 1-year nominal gold return: +0.237 (1971 to 2000), -0.368 (2001 to 2021), -0.804 (2022 to 2024). The 1-year predictive claim has never been positively correlated in a full-era measurement. The pillar is still included because the economic mechanism is coherent; see the limitations section of the methodology page.
1. Method in one paragraph
Monthly resolution, 1971-01-31 to 2026-07-31 (667 points). Nominal price = World Bank Pink Sheet monthly USD per troy ounce; real price = nominal deflated by same-month US CPI-U. Publication lags applied per source (CPI 30 days, COT 3 days, Pink Sheet 7 days, IMF IRFCL 60 days, and so on). At each evaluation date, every pillar is percentile-ranked in its own expanding history to that date; premiums pillar is absent from the backtest (the collector began 2026-08-04). Available pillar weights renormalize to 100 at each date. For each date, forward 1/3/5/10-year nominal and real gold returns are then computed, along with maximum drawdown within each horizon and a DCA versus lump-sum comparison.
2. Buyer-side band statistics
Median outcomes across all monthly observations in each score band. Baseline hit-rate benchmark = all-sample median 1-year nominal return of 5.9%.
| Zone | Band | N | Med 1y nom | Med 3y nom | Med 5y nom | Med 10y nom | Med 1y real | Med 5y real | Med maxdd 1y | Med maxdd 5y | Hit rate 1y | Lump 5y | DCA 5y | DCA vs lump |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Historically very favorable | 80-100 | 0 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a |
| Favorable | 60-79 | 103 | 5.2% | 40.4% | 91.2% | 127.8% | 2.6% | 61.8% | -3.8% | -7.6% | 4854.0% | 91.2% | 83.6% | -2.8% |
| Mixed | 40-59 | 357 | 4.6% | 13.8% | 42.4% | 81.3% | 1.5% | 19.2% | -4.4% | -9.1% | 4762.0% | 42.4% | 42.1% | -1.7% |
| Unfavorable | 20-39 | 174 | 8.6% | 4.0% | 4.9% | 5.5% | 5.7% | -13.9% | -1.8% | -15.2% | 5521.0% | 4.9% | 6.5% | -0.7% |
| Historically very unfavorable | 0-19 | 33 | 9.9% | 21.1% | 64.6% | 42.0% | 6.1% | 22.1% | -5.3% | -11.5% | 5312.0% | 64.6% | 43.9% | -2.9% |
Reading note. Nominal columns include CPI inflation; the real columns are the honest number for wealth preservation. Where the median nominal is positive but the median real is not, inflation ate the gain.
3. Seller-side dual reading
Per the design decision, the instrument reads both ways from a single score. Two seller-relevant numbers are derived from each band: opportunity cost of selling (median forward return forgone), and drawdown avoided by selling (median forward decline avoided). Values are asymmetric by construction: only the historically-observed sign shows up.
| Zone | Band | N | Opp cost sell 1y | Opp cost sell 3y | DD avoided sell 1y | DD avoided sell 3y | Med maxdd avoided 1y |
|---|---|---|---|---|---|---|---|
| Historically very favorable | 80-100 | 0 | n/a | n/a | n/a | n/a | n/a |
| Favorable | 60-79 | 103 | 5.2% | 40.4% | 0.0% | 0.0% | 3.8% |
| Mixed | 40-59 | 357 | 4.6% | 13.8% | 0.0% | 0.0% | 4.4% |
| Unfavorable | 20-39 | 174 | 8.6% | 4.0% | 0.0% | 0.0% | 1.8% |
| Historically very unfavorable | 0-19 | 33 | 9.9% | 21.1% | 0.0% | 0.0% | 5.3% |
4. Where the framework failed
Section title as-is: this is the mandatory section. The three subsections below quantify the failures rather than paper over them.
4.1 Real-rate relationship by era
Pearson correlation between the real-rates sub-score and the subsequent 12-month nominal gold return in three eras.
| Era | N months | Correlation | Interpretation |
|---|---|---|---|
| 1971-01 to 2000-12 | 360 | +0.237 | positive (sub-score points at higher subsequent returns) |
| 2001-01 to 2021-12 | 252 | -0.368 | negative (sub-score points at lower subsequent returns) |
| 2022-01 to 2024-12 | 36 | -0.804 | negative (sub-score points at lower subsequent returns) |
The pillar is not dropped: economic mechanism is coherent and two decades of validity outweigh short breaks. But the disclosure is quantitative, not adjectival.
4.2 The 2022 to 2024 real-rate break, month by month
Excerpt of 36 monthly observations. Full record in the public archive (see the data page).
| Date | Composite | Zone | Real-rates sub-score | Fwd 1y nominal |
|---|---|---|---|---|
| 2022-01-31 | 53 | Mixed | 89.0 | +4.5% |
| 2022-02-28 | 51 | Mixed | 87.5 | -0.1% |
| 2022-03-31 | 49 | Mixed | 86.1 | -1.8% |
| 2022-06-30 | 43 | Mixed | 51.2 | +5.8% |
| 2024-07-31 | 28 | Unfavorable | 19.6 | +39.3% |
| 2024-09-30 | 29 | Unfavorable | 33.5 | +42.7% |
| 2024-10-31 | 24 | Unfavorable | 18.4 | +50.9% |
| 2024-11-30 | 21 | Unfavorable | 17.5 | +54.2% |
| 2024-12-31 | 21 | Unfavorable | 8.4 | +62.7% |
4.3 Band monotonicity check
If the score is well-oriented, medians should rise from Historically very unfavorable to Historically very favorable. Reported violations follow.
- 1-year: Historically very unfavorable 9.87% exceeded Unfavorable 8.6%. Unfavorable 8.6% exceeded Mixed 4.58%. Two violations.
- 3-year: Historically very unfavorable 21.12% exceeded Unfavorable 3.99%. One violation.
- 5-year: Historically very unfavorable 64.59% exceeded Unfavorable 4.89%. One violation.
- 10-year: Historically very unfavorable 42.04% exceeded Unfavorable 5.53%. One violation.
The 33 months in the lowest zone are concentrated in specific historical episodes. Their forward outcomes are not representative of a randomly-drawn "very unfavorable" reading. The n is disclosed here for the same reason.
5. Robustness
5.1 Band edges shifted by 5 points
The canonical bands are 80/60/40/20. Recomputed with edges +5 (85/65/45/25) and -5 (75/55/35/15). The point of the check: does the qualitative story (higher zone -> higher long-horizon median) survive edge shifts?
| Variant | Zone | N | Median 1y | Median 5y |
|---|---|---|---|---|
| edges_+5 | Historically very favorable | 0 | n/a | n/a |
| edges_+5 | Favorable | 39 | 16.5% | 177.6% |
| edges_+5 | Mixed | 353 | 5.0% | 46.7% |
| edges_+5 | Unfavorable | 192 | 6.9% | 4.5% |
| edges_+5 | Historically very unfavorable | 83 | 6.7% | 5.1% |
| edges_-5 | Historically very favorable | 3 | 21.1% | 235.9% |
| edges_-5 | Favorable | 221 | 4.8% | 57.6% |
| edges_-5 | Mixed | 281 | 5.1% | 31.9% |
| edges_-5 | Unfavorable | 139 | 10.2% | 5.1% |
| edges_-5 | Historically very unfavorable | 23 | 11.1% | 124.2% |
5.2 Equal weights instead of design weights
Design weights 25/20/15/10/10/10/10 reallocated to equal per active non-premium pillar. Ordering by band should look qualitatively similar if the score is robust to weight choice.
| Zone | N | Median 1y | Median 5y |
|---|---|---|---|
| Historically very favorable | 0 | n/a | n/a |
| Favorable | 78 | 6.2% | 83.8% |
| Mixed | 393 | 4.6% | 40.4% |
| Unfavorable | 165 | 8.5% | 5.3% |
| Historically very unfavorable | 31 | 9.7% | 116.4% |
6. Era-aware pillar availability
Pillars enter the composite as their data becomes usable. Weights renormalize per era. The premiums pillar is structurally absent from the backtest (collector began 2026-08-04). First active month per pillar in the backtest:
| Pillar | First active month | Months covered |
|---|---|---|
| real_rates | 1971-01-31 | 667 |
| entry_price | 1971-01-31 | 667 |
| structural_demand | 2003-08-31 | 276 |
| dollar | 2010-12-31 | 188 |
| positioning | 2011-05-31 | 183 |
| volatility | 2014-08-31 | 144 |
| premiums | n/a | 0 |
7. Score archive (recent months)
Last 24 monthly readings from the point-in-time backtest and the live engine. The full archive is at /data/history.csv and mirrored at github.com/thegoldbarometer/data.
| Date | Score | Zone | Pillars used | Source |
|---|---|---|---|---|
| 2026-08-05 | 22 | Unfavorable | 6 | live |
| 2026-07-31 | 22 | Unfavorable | 6 | backtest |
| 2026-06-30 | 22 | Unfavorable | 6 | backtest |
| 2026-05-31 | 23 | Unfavorable | 6 | backtest |
| 2026-04-30 | 27 | Unfavorable | 6 | backtest |
| 2026-03-31 | 23 | Unfavorable | 6 | backtest |
| 2026-02-28 | 31 | Unfavorable | 6 | backtest |
| 2026-01-31 | 30 | Unfavorable | 6 | backtest |
| 2025-12-31 | 29 | Unfavorable | 6 | backtest |
| 2025-11-30 | 30 | Unfavorable | 6 | backtest |
| 2025-10-31 | 31 | Unfavorable | 6 | backtest |
| 2025-09-30 | 30 | Unfavorable | 6 | backtest |
| 2025-08-31 | 32 | Unfavorable | 6 | backtest |
| 2025-07-31 | 30 | Unfavorable | 6 | backtest |
| 2025-06-30 | 30 | Unfavorable | 6 | backtest |
| 2025-05-31 | 25 | Unfavorable | 6 | backtest |
| 2025-04-30 | 27 | Unfavorable | 6 | backtest |
| 2025-03-31 | 26 | Unfavorable | 6 | backtest |
| 2025-02-28 | 24 | Unfavorable | 6 | backtest |
| 2025-01-31 | 21 | Unfavorable | 6 | backtest |
| 2024-12-31 | 21 | Unfavorable | 6 | backtest |
| 2024-11-30 | 21 | Unfavorable | 6 | backtest |
| 2024-10-31 | 22 | Unfavorable | 6 | backtest |
| 2024-09-30 | 27 | Unfavorable | 6 | backtest |
8. What this backtest is not
It is not a trading strategy. There are no entry or exit rules, no position sizing, no execution costs, no rebalance frequency. It is a description of what gold did after every historical reading of the instrument. All statistics are unconditional medians and percentile summaries across the bucketed months, drawn from a single 1971-present sample.
It is also not an out-of-sample test. Every date's percentiles are computed from data as of that date only, but the score's orientations (invert real rates, invert dollar level, invert price valuation, etc.) are set by the design based on published economic evidence that has itself accumulated over the same window. That is a known epistemological limit of any single-history backtest of a theory-derived score.
Source note. Nominal gold price series is the World Bank Pink Sheet (CC BY 4.0). Inflation deflator is US Bureau of Labor Statistics CPI-U CUSR0000SA0. LBMA / IBA gold price series are not used, per data-rights constraints (see provenance).