The Ray Dalio All-Weather Portfolio: Allocation and Real Returns

The portfolio

Dalio’s All-Weather balances risk across four economic regimes rather than splitting money between stocks and bonds. The weights below are the retail formulation, and the three figures are forward-looking estimates from J.P. Morgan’s assumptions, not a backtest. Everything after this section is the measured record.

Tony Robbins published a different retail version after interviewing Dalio, the All Seasons portfolio, which holds no inflation-linked bonds and no cash and puts 40% in long Treasuries rather than 15%. It is a separate allocation with a separate record, and the two names get used for each other constantly.

Forward-looking estimates over a long horizon, not a measurement of any past period. Built from J.P. Morgan’s 2026 Long-Term Capital Market Assumptions, annualized and compounded, with a 3.10% risk-free rate. Everything else on this page is the measured record.

Allocation

What if you add Bitcoin?

These are forward-looking estimates rather than a backtest. Nothing below is what Bitcoin did; it is what the portfolio would return if every asset delivered its expected return from here. Each row funds Bitcoin by reducing the other positions proportionally.

The assumption doing the work is Bitcoin at 15.00% a year with 42.5% volatility. That is far below its history and deliberately so: a forward estimate for an asset this young is a judgment, not an extrapolation. It also explains why the returns below move less than people expect. A 10% position in a 15% asset can only add about a point a year to a portfolio, because it is 10% of the portfolio. What changes more than the return is the Sharpe ratio, which is the column worth reading.

Returns are geometric (compound) and forward-looking, built from J.P. Morgan’s 2026 Long-Term Capital Market Assumptions with a 3.10% risk-free rate. Run your own weights in the optimizer.

The headline numbers

Start with the window most comparisons on the internet use, and the one this site used until recently: April 2008 to July 2026. There is nothing meaningful about that start date. It is simply the first month every asset class in all three portfolios has a traded fund to price it from, because the global equity fund launched in 2008. Hold that thought, because it turns out to matter more than anything else here.

Over those eighteen years the All-Weather portfolio returned 5.35% a year. A 60/40 returned 6.85%. Global equities returned 8.65%.

If you came here to find out whether All-Weather beat a conventional portfolio, the answer is no, and it is not close. Eighteen years of compounding left it about $8,000 behind a 60/40 on a $10,000 stake, and about $20,000 behind an all-equity portfolio.

That is the honest headline, and most write-ups of this portfolio never get round to printing it. But it is also close to meaningless on its own, because return was never what the design was optimizing for.

What it was actually built to do

Dalio’s premise is that a conventional portfolio is not balanced at all. A 60/40 holds 60% of its capital in equities, but because equities are roughly three times as volatile as bonds, something like 90% of its risk sits in that one position. It is an equity portfolio with a bond-shaped comfort blanket.

All-Weather balances exposure across economic environments instead: rising growth, falling growth, rising inflation, falling inflation. The version we test holds 30% global equities, 15% long US Treasuries, 15% intermediate US Treasuries, 15% TIPS, 7.5% gold, 7.5% commodities and 10% cash. The long sleeve is the one that matters most and the one this site got wrong until August 2026: bonds only balance equity risk if they carry enough of it, and a bond’s risk comes from its duration. The full allocation, with expected returns and Bitcoin variants, is on our All-Weather portfolio page.

Judged against that objective rather than against a return target, the record is rather good.

The crisis it was designed for

All three portfolios, All-Weather and the 60/40 and all-equity, entered their worst fall in May 2008 and bottomed together in February 2009. What happened after that is the entire argument for the design.

All-Weather lost a fifth of its value and was whole again inside a year. The all-equity portfolio lost more than half and spent nearly four years climbing back. On rolling twelve-month windows the worst stretch All-Weather ever suffered was 15.0%, against 24.2% for the 60/40 and 42.6% for equities.

The best twelve months tell the same story from the other side: All-Weather gained 24.2% at its strongest, against 60.1% for equities. It gives up the spectacular recoveries as well as the collapses. Both extremes are the same design decision.

Why this particular period flatters the alternatives

One important caveat, and it cuts against our own headline table. April 2008 to July 2026 contains the financial crisis and then one of the strongest equity bull markets in history, alongside a difficult stretch for bonds and a mediocre one for commodities.

In other words, this is close to a best case for equity-heavy portfolios and a near-worst case for a design that deliberately spreads risk away from equities. All-Weather did not underperform because the idea failed. It underperformed because the environment it protects against, sustained equity weakness, largely did not arrive after 2009.

Any backtest is a description of one path through history. That is the reason we do not build portfolios from them.

What forward-looking assumptions say

Instead of extrapolating the past, our optimizer runs on published capital market assumptions, estimates of what each asset class should return from today’s starting point. On those numbers the picture changes considerably.

The eighteen-year gap of 1.5 percentage points a year narrows to 0.7. On risk-adjusted terms, All-Weather and the 60/40 are effectively indistinguishable at 0.39 against 0.36, and both are ahead of an all-equity portfolio, which now looks like the weakest of the three per unit of risk rather than the obvious winner.

All-Weather's realised return over this historical sample is 5.35%; its current forward assumption is 5.66%. Equities realised 8.65%, compared with a current forward assumption of 7.00%. These refer to different periods. Comparing a historical outcome with today's forecast is not a test of forecast accuracy; that requires the forecast available at the start of the historical period.

This is not the portfolio Bridgewater runs

Everything above, and every other All-Weather backtest you will find, measures the unlevered version. Bridgewater's actual strategy takes the same risk-balanced mix and levers it to a target volatility, to raise the return and risk of the underlying mix. Balancing risk contributions does not guarantee the highest return per unit of risk; that depends on the assets, their correlations and the period measured.

The buyable version makes the gap concrete. State Street's Bridgewater All Weather ETF, launched 5 March 2025, holds 187% notional exposure: 70.41% global nominal bonds, 42.66% global equities, 40.69% inflation-linked bonds, 33.2% commodities. Those add to nearly twice the money in the fund. The portfolio in the table above adds to one hundred.

Over the fund's own life, leverage helped enormously

The eighteen-year figure is a compound annual rate spanning the financial crisis, 2015, 2018, 2020 and the 2022 bond rout. The fund has existed since 5 March 2025, seventeen months that have been kind to almost everything it holds. Comparing the two directly is the most natural mistake to make with this page, so here is the same portfolio measured through the fund factsheet date, with the first partial month difference stated below.

Levering the same mix to the fund's exposure took 11.90% a year to 18.48%. That is leverage doing exactly what it is supposed to do, in a window where the underlying portfolio was earning well above its financing cost.

Our modeled 18.48% also runs ahead of the fund's own 15.08%, and the gap is the difference between a spreadsheet and a fund: roughly 0.85% a year, plus futures roll costs and leverage that moves rather than sitting fixed. A model that matched a real product exactly would be the suspicious result, not this one.

Over the fund comparison window, leverage increased the modeled compound return and volatility. Its arithmetic Sharpe ratio moved from 1.29 to 1.25after the financing spread. A higher compound return does not by itself mean a better risk-adjusted result.

With a constant exposure and borrowing at the same dated cash benchmark, leverage scales the excess return and its deviation together, leaving the arithmetic Sharpe ratio unchanged before additional costs. A borrowing spread lowers it. Over 2008-04 to 2026-08, unlevered All-Weather's 0.57 was slightly above the 60/40's 0.56, while the leveraged 12% volatility row was 0.55. The measured advantage is small in this window and wider in the longer window below; neither establishes a future advantage.

Two further things leverage does that a return table hides. Volatility scales with the exposure, so the drag on compounding scales with its square: past a point, more leverage compounds worse rather than better. And a single fall of one divided by the exposure ends the position outright. At the exposure the ETF carries, that is a fall of about 53%, against the 51.7% global equities actually delivered inside this very dataset.

You can run this yourself: the optimizer offers a target volatility on the Risk Parity method, and the backtester and simulator both model financing and ruin rather than assuming them away.

Almost all of this depends on starting in 2008

Every figure above begins in April 2008, and not for any good reason: it is simply when the exchange-traded funds we price these assets from began trading. That start date does an enormous amount of work. It opens the record at the financial crisis and then runs through the entire recovery, which is about the most flattering window equities have ever had.

Pricing the same assets from longer-running mutual funds and indices moves the floor back to 2000 09, and 26 years tells a different story. This window includes the dot-com bust and the 2002 bear market: the regime All-Weather exists for, and the one the eighteen-year record leaves out entirely.

And this is the window where leverage earns its keep. At the exposure the Bridgewater fund carries, All-Weather returned 8.33% a year against a 60/40's 5.99%, more return from the design that lost on return over the shorter window. That is the whole argument for risk parity working as advertised: a better ratio, levered until it becomes a better number. It also cost a 32% peak-to-trough fall to get there, against 17% unlevered.

The lesson is not that one of these windows is the true one. It is that an eighteen-year backtest of a strategy built for economic regimes contained essentially one regime, and that every comparison table on the internet, ours included until today, was quietly reporting that.

Fifty-five years, including the decade it was built for

This reconstruction uses the same arithmetic Sharpe definition as the main backtest: average monthly return above the dated cash return, divided by the sample deviation of those differences and annualised by the square root of twelve. Cash comes from this reconstruction's own Treasury-bill series. Charts and tables charge the same financing spread on borrowed exposure; an uninvested balance earns cash.

Everything above stops at 2000, and the reason is unglamorous: it is when the funds we price these assets from start. That leaves out the 1970s, which is a problem, because the 1970s are the whole argument. All Weather exists for the case where growth stalls and inflation runs, and that case has happened once in living memory.

So we rebuilt the data. US equities and cash come from Ken French’s library, which reaches 1926. Gold comes from a fixing series published since 1833. Commodity futures come from AQR’s Commodities for the Long Run dataset, which starts in 1877. Treasury returns are derived from the Federal Reserve’s published yields. International equities come from French’s international index, built on the same MSCI data Bridgewater used, from 1975.

Over 55 years the unlevered portfolio returned 8.29% against a 60/40’s 8.44%, which is roughly a draw on return, from a worst fall of 19.3% against 29.2%. Levered to 12% volatility it returned 10.91%, more than an all-equity portfolio’s 9.57% and from less volatility. That is the risk parity claim stated in full: a better ratio, borrowed against until it becomes a better number.

The 1970s

Isolate the decade and the case stops being subtle.

All Weather returned 11.26% a year through the 1970s and never fell more than 9.3%. A 60/40 returned 6.58% and fell 29.2%. Global equities returned 6.02% and fell 46.5%. Inflation over the same decade ran at 7.39% a year, so both conventional portfolios finished the 1970s poorer than they started in anything that buys goods, and spent about a third of their value getting there. All Weather beat inflation by 3.9 points a year.

What we could not build, and what it changes

One sleeve of the real portfolio has no 1970s equivalent, and it is not a sourcing failure: inflation-linked bonds did not exist in the United States until 1997. Bridgewater hit the same wall and estimated the sleeve backwards, which their own disclosure states plainly.

We tried the same thing and stopped, because the estimate did not survive testing. Modeling an inflation-linked bond needs a real yield, which needs expected inflation, and the usual stand-in for expected inflation before the 1980s is trailing realized inflation. Checked against the Cleveland Fed’s series, that stand-in is almost uncorrelated with the thing it is standing in for: a correlation of 0.11 to 0.18 and an average error of 2.4 percentage points. At a ten-year duration a 2.4 point error in the yield is worth about a quarter of the bond’s price. We would have published a chart made of nothing.

So the 15% inflation-linked sleeve is held in the assets that do its two jobs instead, split between long Treasuries for the duration and commodities for the inflation. That is a judgment call, and it is the one most able to decide the answer, so here is every reasonable version of it rather than the flattering one:

The weakest version of the substitution still returned 9.53% through the 1970s against a 60/40’s 6.58%. Over the full 55 years every version lands between 0.57 and 0.60 on the arithmetic Sharpe ratio, against 0.47 for a 60/40 and 0.41 for equities. The ranking does not depend on the choice. Only the size of the gap does.

Two smaller gaps, for completeness. International equities begin in 1975, so for the first five years that sleeve is held in US equities. And Treasury returns are derived from published yields rather than read off a fund, because the Federal Reserve publishes yields, not returns.

What Bridgewater’s own numbers say

Everything above is our reconstruction, and it keeps landing short of the figure Dalio quotes in interviews. He talks about All Weather compounding at a double-digit rate since inception. Our unlevered version returned 5.35% a year over eighteen years. Something has to explain the gap, and it turns out Bridgewater published the answer.

All Weather is a private fund, so its record is not generally quotable. But Bridgewater presented it to the San Diego County Employees Retirement Association, and because that is a public pension board, the deck went into the agenda papers and stayed there. It is called The All Weather Strategy, 4Q09, and it contains the numbers below.

Their central exhibit runs from 1970 and makes the comparison twice, because the point of risk parity is that you get to choose which end to spend the advantage on. Run All Weather at a conventional portfolio’s risk and you get a much higher return. Run it at a conventional portfolio’s return and you get roughly half the volatility.

13.1% a year is where the remembered figure comes from. It is also the number that needs the heaviest caveat on this page.

The same caveat applies to their longest run, 1925 to 2009, where All Weather simulates to 11.4% a year at a Sharpe ratio of 0.72, against 0.39 for a conventional portfolio and 0.31 for equities. Eighty-four years, thirteen of them real.

Strip the simulation out and a smaller, more useful record is left. From June 1996 to the end of 2009, on live money:

8.4% against a conventional portfolio’s 6.3% is a real advantage and a long way below 13%. The honest summary is that the strategy did beat its benchmark on live money by roughly two points a year over its first thirteen years, and that the double-digit figure belongs to a backtest.

One table in the deck is measured entirely on live money, and it is the one that tests the actual claim. All Weather is sold on holding up when a given market falls. So here is what it did in the quarters each market fell, from the third quarter of 1996 to the end of 2009, net of fees:

In the nineteen quarters the S&P 500 fell by an average of 7.9%, All Weather made 0.2%. In the twenty-four quarters international equities fell, it made 1.1%. That is the design working as described, and it is the part of the record that required no simulation at all.

The deck is from 2009, and it is the most recent one of its kind we can get at. Bridgewater made a similar presentation to San Joaquin County’s retirement board in 2017 covering the record through that August, but it is not posted on the board’s own site and the copies circulating are behind paywalls. So the live-money record we can actually cite stops at the end of 2009. Everything after that is the reconstruction at the top of this page, which is the best anyone outside the fund can do.

Read the primary sources yourself

Almost everything written about All Weather is written about other things written about All Weather. These are the documents underneath it, and every figure on this page that is not our own calculation comes from one of them.

The calculations that are ours are downloadable too: every monthly return from January 1970, each sleeve in its own column alongside the blended portfolio, so any figure on this page can be checked rather than taken on trust.

So should you hold it?

The useful question is not whether All-Weather is better, but which of these three you would actually still own in March 2009.

If you are decades from needing the money and genuinely unmoved by a 50% decline, the historical record says the equity portfolio wins and All-Weather costs you meaningfully. If you are drawing an income, or you know from experience that you sell when things get frightening, the drawdown column matters far more than the return column, and All-Weather has an unusually strong claim.

Most people are somewhere between those poles, which is roughly where the 60/40 sits. None of this makes any of the three wrong. It makes them different answers to a question you have to ask yourself honestly first.

Methodology

Monthly total returns from April 2008 to July 2026, the longest window over which every asset class in all three portfolios has continuous data. Annual rebalancing, no management fee, no transaction costs and no tax. Volatility is the standard deviation of monthly returns annualized; drawdowns are peak-to-trough on month-end values, so intra-month lows were deeper than shown. Forward-looking figures use J.P. Morgan’s 2026 Long-Term Capital Market Assumptions, detailed on our methodology page. Past performance is a record of one sequence of events and not a guide to future returns; nothing here is personal financial advice.