THE SHORT ANSWER
One start date can tell a story; rolling windows show how often that story changed
A rolling-return study applies one fixed holding period to every completed starting month. In this World Bank monthly-average history, short windows vary widely because one entry and exit pair dominates the result. Longer windows narrow some of that endpoint sensitivity, but they still include materially different monetary regimes. The figures are nominal USD benchmark returns before any physical-product or fund costs.
787 completed overlapping windows.
763 completed overlapping windows.
739 completed overlapping windows.
679 completed overlapping windows.
INTERACTIVE ROLLING WINDOWS
Change the horizon, then inspect every completed endpoint
Choose 1, 3, 5, 10 or 20 years. The vertical and horizontal crosshairs snap to an endpoint month and its annualized return. Use pointer, touch or arrow keys; download the selected source-derived series as CSV.
Crosshairs snap to a calculated observation. Monthly averages smooth intramonth highs, lows and drawdowns; this chart is not a daily close series. Historical windows overlap and are not independent forecasts.
Source: World Bank Commodity Price Data (The Pink Sheet), monthly nominal USD per troy ounce, CC BY 4.0. GoldObserve serves the bundled local series; page loads do not download the upstream workbook.
ENDING RESULT × INTERIM PATH MAP
A positive endpoint can still require surviving a deep drawdown
Plot every completed holding window by its annualized endpoint return and its maximum interim drawdown. The two axes answer different questions: where a window finished and how difficult its monthly-average path became before that finish. Select a horizon and starting month to keep those outcomes separate.
Exact 120-month endpoints.
79.7% of overlapping windows.
64.5% of positive endpoints.
How to read it: moving right improves the annualized endpoint result; moving down means the path contained a deeper fall from an earlier monthly-average peak. The horizontal scale changes by horizon so long-window outcomes remain legible, while the fixed vertical scale preserves drawdown comparability. A positive endpoint does not mean the path stayed profitable throughout.
| Holding period | Completed windows | Positive endpoints | Positive + ≥30% drawdown | Share of positive endpoints | Deepest drawdown among positive endpoints | Observed annualized range |
|---|---|---|---|---|---|---|
| 1Y | 787 | 436 | 0 | 0.0% | 1979-05 / -23.9% | -36.4% to +197.4% |
| 3Y | 763 | 478 | 34 | 7.1% | 1979-06 / -53.3% | -15.8% to +72.3% |
| 5Y | 739 | 511 | 124 | 24.3% | 1977-06 / -53.3% | -14.8% to +39.4% |
| 10Y | 679 | 541 | 349 | 64.5% | 1975-02 / -55.7% | -5.6% to +34.4% |
| 20Y | 559 | 507 | 507 | 100.0% | 1979-05 / -59.1% | -4.4% to +15.9% |
Source: World Bank monthly-average nominal USD gold. Endpoint returns and interim drawdowns are GoldObserve calculations from the locally versioned observations.
HORIZON COMPARISON
Median, best and worst annualized results by holding period
| Holding period | Completed windows | Median annualized | Worst annualized | Worst window | Best annualized | Best window |
|---|---|---|---|---|---|---|
| 1 year | 787 | +2.9% | -36.4% | 1980-07 to 1981-07 | +197.4% | 1979-01 to 1980-01 |
| 3 years | 763 | +3.2% | -15.8% | 1980-10 to 1983-10 | +72.3% | 1977-01 to 1980-01 |
| 5 years | 739 | +4.6% | -14.8% | 1980-01 to 1985-01 | +39.4% | 1969-12 to 1974-12 |
| 10 years | 679 | +6.4% | -5.6% | 1980-07 to 1990-07 | +34.4% | 1970-01 to 1980-01 |
| 20 years | 559 | +7.5% | -4.4% | 1980-09 to 2000-09 | +15.9% | 1960-01 to 1980-01 |
REPRODUCIBLE METHOD
Every window uses exact monthly endpoints and compound annualization
Convert 1, 3, 5, 10 or 20 years into 12, 36, 60, 120 or 240 months.
Pair each starting monthly average with the observation exactly that many months later.
(Ending average / starting average - 1) x 100.
[(Ending average / starting average)^(12 / months) - 1] x 100.
Annualization makes horizons comparable, but it does not mean the price grew smoothly each year. A window can contain deep interim losses even when the endpoint return is positive. Inspect gold drawdown history before reducing a path to one rate.
INTERPRETATION BOUNDARY
Overlapping history is evidence, not a bag of independent trials
A January-to-January five-year window and the adjacent February-to-February window share 59 of 60 monthly intervals. Counting them expands endpoint coverage but does not create independent experiments. That dependence is why GoldObserve reports historical frequency and sample extremes without confidence intervals or a claimed probability of future profit.
The early sample also crosses different regimes: the official-price and Bretton Woods era, post-1971 market liberalization, high inflation, disinflation, financial-product growth and later monetary-policy cycles. A long sample is valuable precisely because it is heterogeneous; it should not be summarized as one timeless expected return.
Source and study limitsWorld Bank monthly history · CC BY 4.0
SOURCE, LICENSE & LIMITS
The study uses one consistent monthly history
The source is the World Bank Commodity Price Data (The Pink Sheet), licensed CC BY 4.0. It contains 799 monthly gold observations from 1960-01 through 2026-07. The source workbook was published 2026-08-04, retrieved 2026-08-07, and preserved with SHA-256 7902a77505ebdc5d202ce65f666c2ee1b04b626f042d7738ed3e6f7d112c8433.
The values are nominal US-dollar monthly averages per troy ounce. They are not daily closes, dealer quotes, London auction prices or exact transaction prices.
CONTINUE THE RESEARCH
Separate holding-period evidence from a personal investment result
FREQUENTLY ASKED QUESTIONS
Questions about gold holding-period evidence
What is a rolling gold return?
A rolling return repeats the same holding-period calculation from every eligible starting month. A five-year study therefore compares each monthly average with the monthly average exactly 60 months later.
Are these returns annualized?
Yes. The chart annualizes each exact endpoint ratio using 12 divided by the selected number of months. The table also identifies the best and worst completed windows.
Do overlapping windows count as independent observations?
No. Adjacent rolling windows share most of their months, so the results are descriptive paths rather than independent trials suitable for a simple probability model.
Do the returns include physical gold costs?
No. Premiums, spreads, storage, insurance, tax, financing and product expenses are excluded. Those costs can materially reduce an investor return.
Can rolling returns predict the next holding period?
No. They reveal how endpoint results varied in this historical monthly-average sample. They do not estimate a future expected return.