Key takeaways
- From 1980 to 2025, an annually rebalanced 25/25/25/25 split returned 7.44% a year against 10.24% for a 60/40. That turned $100 into $2,708 rather than $8,854.
- The split's worst calendar year in 46 years was a loss of 8.31%, in 2022, against 17.95% for the 60/40 and 36.55% for all-equity in 2008.
- It beat the 60/40 in two of five rate regimes: by 3.72 points a year from 2000 to 2008, and by 3.96 points from 2022 to 2025.
- Replacing the 25% gold sleeve with cash cost 0.46 points a year and produced a slightly smaller worst year, at a loss of 7.92%.
- Gold returned -2.80% a year from 1980 to 1999. Since 1900 its real return has been 1.3% a year, on the Yearbook's figures.
The permanent portfolio cost 2.80 points a year and bought a much smaller worst year
You want to know whether Harry Browne's four-way split has held up, and what it costs against a plain 60/40.
Here's the arithmetic. From 1980 to 2025, a portfolio holding 25% US stocks, 25% Treasury bonds, 25% gold and 25% Treasury bills, rebalanced back to those weights every year end, returned 7.44% a year. A 60/40 built from the same stock and bond series returned 10.24%. Over 46 years that gap compounds hard. $100 became $2,708 in the four-way split and $8,854 in the 60/40.
What the split bought sits at the other end of the distribution. Its worst calendar year in the whole sample was a loss of 8.31%, in 2022. The 60/40 lost 17.95% in the same year. The standard deviation of the split's annual returns was 6.85%, against 10.69%.
So it's a trade at a stated price, not a free lunch and not a failure. The interesting part is that the price wasn't constant. It moved with the rate regime, and twice it went the other way.
What Browne specified, and what this test substitutes for it
Browne's own bibliography dates the plan precisely: "Inflation-Proofing Your Investments (with Terry Coxon, 1981). The introduction of the Permanent Portfolio plan. The investments suggested have been superseded by the latest version of Fail-Safe Investing." So the author himself treated the 1981 holdings as out of date. What survived is the structure. His own statement of the idea is Rule 11 of his 16 Golden Rules of Financial Safety: "set up a simple, balanced, diversified portfolio. I call this a 'Permanent Portfolio' because once you set it up, you never need to rearrange the investment mix even if your outlook for the future changes."
The four legs aren't generic, and his own results page says so: "Stock results are for an S&P 500 Index mutual fund, and include reinvestment of dividends. Bond results are for a 30-year T-bond, and include interest received. Gold results are for American Eagle 1-ounce coins. Cash results are for Treasury bills, assuming that a 1-year bill was bought at the start of each year."
This test uses Aswath Damodaran's NYU Stern return series for all four legs, and two of them differ from Browne's. The bond leg here is a 10-year Treasury rather than a 30-year, so it moves less in both directions. The cash leg is a 3-month bill rather than a 1-year one. Those substitutions matter, and how much shows up further down.
Four allocations, 46 years, one table
Everything below is annual data, rebalanced once a year, before costs and before tax.
| Allocation | Annualised | After inflation | Worst year | Volatility | $100 grew to |
|---|---|---|---|---|---|
| 100% S&P 500 | 12.11% | 8.65% | -36.55% (2008) | 16.29% | $19,210 |
| 60/40 stocks and bonds | 10.24% | 6.84% | -17.95% (2022) | 10.69% | $8,854 |
| 25/25/25/25 four-way split | 7.44% | 4.12% | -8.31% (2022) | 6.85% | $2,708 |
| 25/25/50, cash in place of gold | 6.98% | 3.68% | -7.92% (2022) | 5.60% | $2,233 |
US consumer prices rose 3.18% a year over the same 46 years, which is the whole distance between the first two columns. Note the ordering. Return and volatility fall together down the table, and the worst year falls with them. Nothing in the four-way split beats the 60/40 on this view. It just loses less when it loses.
The gap moves with the rate regime, and it flips twice
Split the sample at the turns in the Federal Reserve's own policy rate, published in the H.15 release. The effective funds rate averaged 16.39% in 1981 and peaked at a monthly average of 19.10% in June that year. It reached 0.08% in 2021 and 4.21% in 2025. Those are not variations on one regime. They're different worlds, and the four-way split behaves differently in each.
| Regime | Four-way split | 60/40 | Gap |
|---|---|---|---|
| 1980-1981, rates at their peak | 4.66% | 8.80% | -4.14 pts |
| 1982-1999, the long disinflation | 8.70% | 15.34% | -6.64 pts |
| 2000-2008, two equity bears | 5.78% | 2.06% | +3.72 pts |
| 2009-2021, rates near zero | 6.59% | 10.84% | -4.25 pts |
| 2022-2025, the reset | 9.70% | 5.73% | +3.96 pts |
The chart above plots that last column. Three regimes negative, two positive, and the full-period average of -2.80 points is a blend of two quite different behaviours rather than a description of either.
The two regimes it won were the two when gold worked
2000 to 2008 and 2022 to 2025 don't look alike from the inside. In the first, US stocks fell 3.56% a year while 10-year Treasuries returned 8.39%. Bonds did their job, and the 60/40 still made only 2.06%, because the 60 was the problem. In the second, bonds were the problem, losing 2.46% a year while stocks made 11.02%.
What the two windows share is gold. It compounded at 12.97% a year through the first and 24.26% through the second, against -1.73% a year in the 1982-1999 disinflation. At a quarter weight, that single asset is the whole margin.
2022 is the sharpest year in the sample. US stocks lost 18.04% and 10-year Treasuries lost 17.83% in the same calendar year, so the diversification inside a 60/40 did nothing at all. Gold returned 0.55% and Treasury bills 2.09%. In the 46 years from 1980, only 2018 and 2022 saw both stocks and bonds finish negative.
2025 is the other extreme. Gold returned 66.22%, contributing 16.56 points on its own and pulling the four-way split to 23.99% for the year, its best in the sample, against 13.75% for the 60/40.
The 25% gold sleeve is the strongest objection to the design
Here's the case against, and it's a serious one.
Gold returned -2.80% a year from 1980 to 1999, and -6.54% a year after inflation. Two decades of losing money in real terms, at a quarter of the portfolio. Browne said as much himself in Rule 10: "gold and silver became the losers of the 1980s and 1990s, while stocks and bonds multiplied their value."
The long-run picture is no kinder. The 2026 Global Investment Returns Yearbook, compiled by Elroy Dimson, Paul Marsh and Mike Staunton, reports that the real US dollar gold price has risen 5.2-fold since 1900, "an annualized return of 1.3%". On the inflation-hedge claim specifically, the Yearbook is blunt: "Of the 28 years in which inflation exceeded 3%, we find that gold returns were negative in 13 of them." A hedge that lost money in nearly half the high-inflation years on record is a weak hedge. Measured the same way, bitcoin real returns beat US inflation in only 10 of the 27 months when annual CPI ran at 4% or more.
The swap test in the table above sharpens it. Move the 25% gold to cash and the portfolio returned 6.98% instead of 7.44%, so gold added 0.46 points a year. But the worst year got marginally better without it, at -7.92% against -8.31%, and volatility fell from 6.85% to 5.60%. Over these 46 years the gold sleeve paid in return, not in protection. Nearly all of that return came from the two regimes the split won: gold compounded at 12.97% from 2000 to 2008 and 24.26% from 2022 to 2025, against -1.73% through the 1982-1999 disinflation. If you want the wider evidence on how much gold in a portfolio the drawdown record supports, it's a separate question from this one.
Browne's own numbers don't match this recomputation
His results page, which runs from 1970 through 2003, lists 1980 at +22.1%, 1981 at -6.2% and 1982 at +23.3%. This series gives +13.83%, -3.77% and +19.99% for the same three years. That's a wide gap, and it isn't a disagreement about what happened.
It's a disagreement about instruments. Three of the four legs are measured differently: a 30-year Treasury against a 10-year, a 1-year bill against a 3-month one, and coin prices against an annual gold price series. Every published backtest of this strategy is really a backtest of one particular set of proxies. That is worth holding on to when a number gets quoted at you without its footnote.
The fund carrying the name isn't the portfolio
Permanent Portfolio, the mutual fund, has run since its Class I shares launched on 1 December 1982. Its prospectus sets six fixed target percentages: "Gold 20%, Silver 5%, Swiss franc assets 10%, Real estate and natural resource stocks 15%, Aggressive growth stocks 15%, Dollar assets 35%." They are "fundamental and cannot be changed without shareholder approval". That's not the four-way split, and it never was.
Its record is a useful reality check on cost. Class I returned 7.07% a year before taxes from inception to the end of 2025, against 12.05% for the S&P 500 over the same span. In 2025 alone it returned 28.78%. Total annual operating expenses run 0.79% for Class I and 1.79% for Class C, which is a meaningful bite out of a 7% gross number.
What this backtest cannot tell you
It's one path through history, and a specific one. It spans the whole range of the Federal Reserve's 10-year Treasury series: the highest annual average in that series is 13.92%, in 1981, and the lowest is 0.89%, in 2020. Four decades of falling yields flattered both portfolios, and flattered the 60/40 more, because it carries a 40% bond weight against the split's 25% and holds nothing that gains when bonds lose.
The figures are before costs and before tax. Rebalancing a quarter of a portfolio into and out of gold every year is a taxable event in a taxable account, and the cheapest share class of the fund that does something similar charges 0.79%. The sample is also unbalanced. 2022 to 2025 is four years and 1980 to 1981 is two, against 18 years for the disinflation window, so the two regimes the split won carry much less data than the two it lost.
It is entirely US, in dollars. The Yearbook covers 35 markets and finds developed-market equities returned 8.5% a year since 1900 against 6.9% for emerging markets, so a sterling or euro version of this test would not produce these numbers. And the proxies aren't Browne's, as the section above sets out. A backtest is a measurement of the past under assumptions, not a forecast.
What would change the conclusion
If the 1981 to 2020 bond repricing doesn't recur. The 10-year yield went from 13.92% to 0.89% and back to 4.29% in 2025. A large part of the 60/40's 6.64-point annual margin through the disinflation came from that one-way move, and a yield can only fall from 13.92% to 0.89% once.
If gold's next stretch looks like its first two decades here. At -2.80% a year from 1980 to 1999, the sleeve that produced the split's edge in 2000-2008 and 2022-2025 was the sleeve that produced its deficit before that. The same 25% weight does both jobs.
If stocks and bonds stop falling together. In 46 calendar years only 2018 and 2022 saw both finish negative, and 2022 does most of the work in the window the four-way split won most convincingly. Two observations is not a base rate. A backtest can price what happened in 2022. It can't price how often that recurs, and that frequency is what the whole four-way case rests on.
What's measurable meanwhile is your own drift. Gold's annual returns had a standard deviation of 17.98% across this sample, wider than the S&P 500's 16.29%, so the sleeve carrying the argument is also the one that pulls furthest from its target between rebalancing dates. LedgerTouch shows the four weights against their targets continuously, which is the number this whole argument is really about. The all weather portfolio and cash as an asset class are the two nearest tests to this one, and neither settles it either.