Gold Allocation Backtest: What Happened When 5%–25% Gold Was Added to a 60/40 Portfolio

Educational only: All metrics on this page are historical, period-dependent, and illustrative. Past performance does not guarantee future results. No allocation shown here is presented as "optimal" for any individual. Customers should speak with a qualified financial or tax advisor before acting on any allocation takeaway. Goldco does not offer tax or legal advice.

What this backtest is, and is not. This backtest models gold as an asset allocation. It does not model the net return of a Gold IRA, which can include dealer spreads, custodian fees, storage fees and other account-specific costs.

1. What we tested

Six portfolios. One variable.

Each portfolio holds a share of gold; the rest is invested in a 60/40 stock and bond mix. The 60/40 ratio is preserved within the non-gold portion, so a 10% gold portfolio is 54% stocks, 36% bonds and 10% gold — not 60/30/10. This matters: it isolates the effect of adding gold, rather than confusing it with a shift in the stock-to-bond balance.

Every portfolio is rebalanced back to its target weights once a year, at each calendar year-end.

2. Data and methodology

All four inputs are public primary sources.

Conventions. Monthly observations, April 1968 to June 2024, 674 return months. Annual rebalancing at calendar year-end. Dividends reinvested. Bond returns independently constructed from Treasury yields, then validated against Aswath Damodaran's published series. Risk-free rate converted from the 3-month bill's discount basis to a bond-equivalent yield, then to monthly.

Costs. None. No transaction costs, dealer spreads, custodian fees, storage, depository charges or taxes are modelled. This is an asset-allocation study, not a simulation of Gold IRA net returns.

The end date is June 2024 because that is where the dividend series ends. The gold data runs to December 2025, but extending only one input would break the like-for-like comparison.

The full technical methodology, validation log and calculation code are linked in section 11.

3. Full-period results (nominal)

Different allocations lead on different measures. Volatility was lowest at 15%. Maximum drawdown was smallest at 20%. Risk-adjusted return, measured by Sharpe ratio, was highest at 20%.

On return, 25% gold produced the numerically highest full-period CAGR at 9.3996%, versus 9.3936% for 20% gold — a difference of only 0.006 percentage points. That difference is too small to support a claim that 25% was meaningfully superior. At the two-decimal presentation used in the table both round to 9.40%.

Every portfolio's worst calendar year was 2008, and the size of that loss fell steadily as the gold share rose, from −15.8% at 0% gold to −11.5% at 25%.

4. Inflation-adjusted results

Nominal figures describe how the account balance grew. Real figures describe what that balance could buy. Over this period prices rose by a factor of 9.1330, so the two tell different stories.

The nominal and real results identify different allocations. In nominal terms the smallest maximum drawdown was at 20% gold. In purchasing-power terms it was at 15% (−23.6%), with 20% slightly worse (−24.8%) and 25% worse again (−27.5%).

The two measures differ because a nominal drawdown counts only the fall in account value, while a real drawdown also counts inflation eroding purchasing power during the same window. A portfolio can decline less in dollar terms yet lose more in what those dollars buy.

For gold's behaviour against inflation directly, rather than inside a portfolio, see our gold versus inflation dataset.

5. Two different questions: one path, or many starting points

The full-period figures in section 3 describe one historical path: an investor who started in April 1968 and held until June 2024. Rolling windows ask a different question — what happened across many different starting points within that same history.

Neither measurement invalidates the other. They are answers to different questions, and the answers differ. A single realised path tells you what actually happened to one investor. A distribution of overlapping windows tells you how varied the experience was depending on when someone began. Both are reported here because reporting only one would misrepresent the record.

10-year rolling windows (555 overlapping windows)

20-year rolling windows (435 overlapping windows)

This is the most important comparison on the page. Over the full period, CAGR rose slightly as gold increased. Across rolling windows the median CAGR moves the other way: it falls as gold increases, from 9.56% to 9.14% over 10-year windows and from 9.91% to 8.83% over 20-year windows.

Both results are correct, and each answers its own question. The full-period figure is one realised path; the rolling medians describe the central tendency across many starting points. Quoting either one alone would give an incomplete picture of the record, which is why both appear here.

Two further observations. The worst 10-year outcome improved steadily with gold, from 1.41% at 0% to 4.29% at 25%, while the best outcome fell. And median drawdown improved with gold in both window lengths.

6. How often did gold reduce drawdown?

For each rolling window, the maximum drawdown of each gold portfolio was compared with the 0% gold portfolio over exactly the same window.

Gold reduced maximum drawdown versus the 0% portfolio in all 435 overlapping 20-year windows examined.

Those windows overlap substantially. Consecutive 20-year windows differ by a single month and share 239 of their 240 observations, so these are not 435 independent observations and the result must not be read as a probability of future success. It describes what happened in this one historical record, examined from many overlapping vantage points.

The same caution applies to the 10-year figures, where roughly 92–93% of the 555 overlapping windows showed reduced drawdown.

For how gold itself behaved during the equity bear markets that produced many of these drawdowns, see our gold versus S&P 500 bear-market dataset, which measures both assets over identical peak-to-trough periods.

7. Start date matters

The identical methodology was rerun from later starting points, all ending June 2024. Nothing else changed.

CAGR by start date

Volatility by start date

Changing the starting date can reverse the apparent return advantage of gold.

Starting in 1968 or 2000, CAGR rose as the gold share rose. Starting in 1980 or 1990, it fell — and 0% gold produced the highest return in two of the five start dates tested. Starting in 1973 the highest CAGR was at 20%, with the relationship neither consistently rising nor falling.

The risk results were far more stable. Every non-zero allocation had lower volatility than 0% gold at every one of the five start dates. That relationship was not uniform, though: in three of the five the lowest volatility occurred at an intermediate allocation rather than the largest one — at 15% for the 1968 and 1973 starts and 20% for 1980, and only at 25% for the 1990 and 2000 starts.

The ordering of CAGR by gold allocation changes materially with the starting date. The 1968 and 2000 results, where returns rose with the gold share, are two specific starting points and are not evidence of a general tendency for gold to increase returns — the 1980 and 1990 starts produced the opposite ordering over periods of 44 and 34 years.

This is the clearest evidence on the page that gold's historical effect on risk was more consistent than its effect on return.

8. Rebalancing matters

The primary results assume the portfolio is rebalanced to target once a year. This section removes that assumption entirely and lets each portfolio drift for 56 years.

A portfolio left untouched for decades becomes a different portfolio. Every one of the six drifted heavily toward equities, ending between 83.4% and 93.3% in stocks, with bonds falling to roughly 6% in every case. Gold ended well below its starting weight in each portfolio.

The consequences are visible in every column. Without rebalancing the return ordering reverses: 0% gold produced the highest CAGR (9.62%) and 25% the lowest (9.27%), the opposite of the rebalanced result. Volatility rose for every portfolio. Maximum drawdown improved from −36.1% at 0% gold to −33.0% at 15%, then worsened to −37.1% at 20% and −40.1% at 25%.

Rebalancing is a material assumption in this study, not a minor methodological detail. It is one of the few inputs capable of reversing the direction of the headline result, and section 3's figures should be read as describing annually rebalanced portfolios rather than gold allocations in general.

This is a sensitivity test, not a strategy. It is shown because the discipline of periodically selling what has risen and buying what has fallen is inseparable from the outcome reported above.

9. Which allocation performed best depends on the measure

There is no single answer, because the allocations that led on each metric were not the same ones.

Note on the CAGR row: at full precision 25% gold returned 9.3996% and 20% returned 9.3936%, a difference of 0.006 percentage points. Both round to 9.40%. That gap is too small to distinguish the two, and neither is presented as the return-maximising allocation.

No allocation appears in every row. The allocation that minimised volatility was not the one that minimised drawdown; the one with the best risk-adjusted return was not the one with the highest raw return; and the allocation with the best rolling-period return was the one holding no gold at all.

Which of these matters depends entirely on what is being optimised — long-run growth, year-to-year stability, the depth of the worst decline, protection of purchasing power, or return per unit of risk. Those are different objectives, and this data does not rank them.

Nothing on this page is a recommendation. These are historical measurements of six specific portfolios over one specific period, computed without costs or taxes. An allocation decision depends on individual circumstances, and should be discussed with a qualified financial or tax advisor.

10. Limitations

On the data and method

• Historical results cannot establish future performance. This is a record of what happened, not a forecast.
• Drawdowns are measured from monthly observations and do not capture intramonth peaks or troughs. Actual worst-case declines within a month would have been deeper.
• The gold series is a monthly average of daily quotations, not a month-end price. Averaging slightly dampens measured volatility and drawdown. The Shiller stock price series is also a monthly average, so the two are consistent with each other.
• Shiller's monthly dividends are interpolated from quarterly totals, so monthly equity returns are approximate even though annual totals are sound.
• The Treasury total-return series is independently constructed from Federal Reserve yields rather than taken from a commercial index. It was validated against Damodaran's published series (correlation 0.991, mean absolute annual difference 0.81 percentage points).
• Rolling windows overlap heavily and are not independent observations.
• Results depend on the start and end date, as section 7 demonstrates directly.
• The rebalancing assumption materially affects the outcome, as section 8 demonstrates.

On what is excluded

• No transaction costs.
• No precious-metals dealer spread between buying and selling price.
• No Gold IRA custodian or administration fees.
• No storage or depository fees.
• No taxes of any kind.
• No required minimum distributions or account-specific cash flows.
• No account minimums, and no assumption about how gold is actually held.

The most important limitation. This study models gold as an asset allocation. It is not a simulation of Gold IRA performance. A real Gold IRA involves dealer spreads, custodian fees, storage costs and distribution rules that are not represented here, and each of those reduces net returns. The gap between an allocation study and a real account is significant.

For the account costs this study excludes, see our Gold IRA fees benchmark, which documents setup, annual administration, storage and total cost by account size from published schedules.

11. Reproducibility

Every figure on this page can be recomputed from public data.

• Methodology specification — conventions, formulas and construction rules
• Validation log — every check run before results were accepted, including two that failed first and the corrections made
• Allocation results (CSV)
• Rolling-window results (CSV)
• Start-date sensitivity (CSV)
• Normalised monthly source series (CSV)
• Source checksums — SHA-256 for every input file
• Calculation code: ingest, bond returns, T-bill returns, backtest, extended analysis, reconciliation

Dataset cutoff: source data downloaded 2 September 2026; analysis period April 1968 to June 2024.

The validation log records that the Treasury construction failed its first external validation with a systematic bias, was diagnosed as a coupon double-count, corrected, and revalidated; and that a portfolio identity check failed on a fault in the test harness rather than the calculation. Both are documented rather than removed, because a validation process that never reports a failure is not evidence of correctness.

Data sources: World Bank Commodity Markets "Pink Sheet"; Robert Shiller, Yale University; Federal Reserve Bank of St. Louis (FRED); Aswath Damodaran, NYU Stern (validation reference). Raw source files are not redistributed here; the retrieval code and checksums above identify the exact files used.

This study is part of our Research Center, where every dataset is published with its methodology, sources and limitations.

Conclusion

In this 1968–2024 historical test, gold's effect on portfolio risk was substantially more consistent than its effect on portfolio return.

The risk finding held up across every way it was measured. Volatility and maximum drawdown generally improved at moderate allocations, every non-zero allocation reduced volatility relative to holding no gold at all five start dates tested, and median drawdown improved with gold across both rolling window lengths.

The return finding did not hold up. Full-period CAGR rose slightly as the gold share increased, but that ordering reversed for two of the five start dates and reversed again across rolling windows, where the portfolio holding no gold produced the highest median return over both 10-year and 20-year periods. A result that changes direction depending on when the measurement starts is not a stable finding.

The annual rebalancing assumption is doing material work. Removed entirely, the portfolios drifted to roughly 83–93% equities and the full-period return ordering reversed. These results describe annually rebalanced portfolios, not gold allocations in the abstract.

No single allocation led on every metric. The allocation with the lowest volatility was not the one with the smallest drawdown, and neither was the one with the highest rolling-period return.

Frequently asked questions

Does adding gold always improve returns?

No. In this backtest the effect on return depended heavily on when the measurement started. Over the full April 1968 to June 2024 period, CAGR rose slightly as the gold share increased. Starting in 1980 or 1990 the ordering reversed and the portfolio holding no gold returned the most. Across rolling 10-year and 20-year windows, the median return was also highest with no gold. The effect on risk measures was more consistent than the effect on returns.

Which gold allocation performed best?

Different allocations led on different measures, so there is no single answer. Volatility was lowest at 15%, maximum drawdown was smallest at 20%, the smallest inflation-adjusted drawdown was at 15%, and the highest median rolling return was at 0%. The full-period CAGR figures for 20% and 25% differ by 0.006 percentage points and cannot be meaningfully distinguished. Which measure matters depends on what an individual is trying to achieve, and nothing here is a recommendation.

Did gold reduce drawdowns?

In this dataset, yes, consistently. Gold reduced maximum drawdown versus the 0% gold portfolio in all 435 overlapping 20-year windows examined, and in roughly 92–93% of the 555 overlapping 10-year windows. Those windows overlap heavily — consecutive 20-year windows share 239 of their 240 monthly observations — so they are not independent trials and the result is not a probability of future success.

How much does the rebalancing assumption matter?

Materially. The primary results assume rebalancing to target once a year. Without any rebalancing, the same six portfolios drifted to roughly 83–93% equities over 56 years, volatility rose for every portfolio, and the full-period return ordering reversed. These results describe annually rebalanced portfolios rather than gold allocations in general.

What data underlies this backtest?

Four public primary sources: gold prices from the World Bank Commodity Markets "Pink Sheet" monthly series; US equity prices and dividends from Robert Shiller's dataset at Yale University; inflation from the Shiller CPI series, verified identical to the Federal Reserve's CPIAUCNS; and Treasury yields from FRED at the Federal Reserve Bank of St. Louis. The Treasury total-return series was constructed independently from those yields and validated against Aswath Damodaran's published series. All figures on this page are computed from these sources. The full methodology, validation log, source checksums and calculation code are published above.

Does this show what a Gold IRA would have returned?

No. This is an asset-allocation study. It excludes dealer spreads, custodian and administration fees, storage and depository charges, transaction costs, taxes and required minimum distributions. A real Gold IRA carries those costs and its net return would be lower than the figures shown here.

Reviewed and edited by Daniel M. — editor, 401kToGoldIRA.org.

Further Reading

How Much Gold Should Investors Own in Retirement?The allocation framework behind the backtest numbers.Gold IRA vs Gold ETF vs Physical GoldOnce you’ve picked an allocation, how to hold it — after-fee comparison.Gold IRA CalculatorModel a diversified vs all-stock scenario with your own assumptions.