Key takeaways
- Bridgewater's published All Weather diagram puts 25% of portfolio risk in each of four economic environments. It names the asset classes and publishes no weights at all.
- Six US-listed ETFs weighted to hit the risk shares that diagram implies come out at 12.0% equities and 42.2% inflation-linked bonds, on volatilities to 30 June 2026.
- That stack returned 2.04% a year over 2021 to 2025, against 8.07% for a 60/40 built from the same funds. The gap is 6.03 percentage points a year.
- It fell 15.06% in 2022, just 1.91 points less than the 60/40's 16.97% loss, in exactly the inflation shock the design exists to survive.
- RPAR, the levered risk parity ETF that does hold 25% of risk in each of four asset classes, lost 22.81% in 2022 and compounded at 0.89% a year over the same five years.
The thing you are trying to copy is a risk diagram, not a portfolio
You want to rebuild Ray Dalio's All Weather strategy out of ETFs, and you want to know how far off you'd land. The awkward part is that there is no published portfolio to land near.
Bridgewater set the concept out in January 2012, in a paper called The All Weather Story. It describes a two-by-two grid. Growth and inflation, each either rising or falling relative to what markets already expect. Every quadrant carries "25% of risk", and the paper maps asset classes onto them: equities, commodities, corporate credit and emerging market credit for rising growth; inflation-linked bonds, commodities and emerging market credit for rising inflation; nominal bonds and inflation-linked bonds for falling growth; equities and nominal bonds for falling inflation.
What the paper never gives is a weight. Not one portfolio percentage, anywhere in nine pages. It doesn't give a return series either. Bridgewater launched the strategy in 1996 for Dalio's own trust assets, and the paper reports no results for it.
So the measurable gap here isn't tracking error against a live benchmark. It's the distance between a rule about risk and the portfolio you can actually assemble in a dealing account. That distance turns out to be large, and most of it is one missing tool.
Turning four boxes into six ETF weights
Start with the only quantity Bridgewater publishes: 25% of risk per box. Split each box evenly among the assets it names, and you get an implied risk budget per asset class. Rising growth names four assets, so each takes 6.25%. Rising inflation names three, so each takes 8.33%. The two remaining boxes name two assets apiece, so each of those takes 12.5%.
Add up what each asset collects across the boxes it appears in, and the risk budget looks like this: nominal bonds 25.0%, inflation-linked bonds 20.8%, equities 18.8%, commodities 14.6%, emerging market credit 14.6%, corporate credit 6.2%.
That even split inside each box is a reading, not Bridgewater's arithmetic. The paper names the assets and stops. Anyone reconstructing this has to invent the same step, which is the first place two honest replications diverge.
Turning a risk share into a pound weight needs a volatility for each sleeve. The iShares factsheets publish a three-year standard deviation, all dated 30 June 2026, so the whole table comes from one vintage. Weight is then risk share divided by volatility, normalised to 100%. The lower the volatility, the bigger the cheque, which is the mechanism the whole idea rests on.
| Box-implied role | Fund | Risk share | 3y volatility | Weight |
|---|---|---|---|---|
| Nominal bonds | TLT | 25.0% | 13.74% | 15.2% |
| Inflation-linked bonds | TIP | 20.8% | 4.12% | 42.2% |
| Equities | IVV | 18.8% | 13.05% | 12.0% |
| Commodities | GSG | 14.6% | 20.42% | 6.0% |
| Emerging market credit | EMB | 14.6% | 6.86% | 17.8% |
| Corporate credit | LQD | 6.2% | 7.65% | 6.8% |
A 12.0% equity weight is the number people stall on. It isn't a view about equities. It falls straight out of the fact that TIP moved 4.12% a year and IVV moved 13.05%, so a pound of TIP buys less than a third as much risk. Balancing risk without borrowing means the calm assets get most of the money.
What an all weather portfolio in ETFs actually returned, 2021 to 2025
Rebalance that stack once a year and run it through the calendar-year returns on the same factsheets. It made 6.84% in 2021, lost 15.06% in 2022, then made 7.35%, 4.25% and 8.92%. Compounded, that's 2.04% a year and 10.62% over the five years.
A 60/40 of IVV and IEF, the same fund family and the same window, returned 8.07% a year and 47.41% cumulatively. IVV on its own compounded at 14.40%. The chart above puts the three side by side with RPAR, the one retail fund that implements the concept properly.
So the all weather portfolio you can build unlevered gave up 6.03 percentage points a year against a plain 60/40. That is the tracking gap, and it is not subtle.
The risk side of the ledger did work as designed. Summing each sleeve's standalone risk, equities supply 18.7% of the stack's risk budget against 75.0% in the 60/40. Bridgewater's paper makes the same observation about institutional portfolios: one with "roughly 60% of its dollars invested in equities" ends up with "almost all of its risk" there. Our own risk contribution workings, which use a full covariance matrix rather than standalone volatilities, land in the same neighbourhood.
The 2022 test, and how little the balance bought
2022 is the year this design is supposed to earn its keep. It's the only year of the five when equities and both kinds of bond fell together, and the only one when a single sleeve finished in the black.
That sleeve was commodities. GSG returned 24.09% while IVV lost 18.13%. Everything meant to cushion the fall fell with it: TIP lost 12.13%, TLT lost 31.41%, LQD lost 18.01% and EMB lost 18.03%. Nominal bonds and inflation-linked bonds together carry 45.8% of the box-implied risk budget, and both of them went down.
The stack finished 2022 down 15.06%. The 60/40 finished down 16.97%. Balancing across four economic environments bought 1.91 percentage points of relief in the one year the framework was built for, and cost 6.03 points a year across the window. Anyone selling all weather investing as crash protection is selling the 1.91 and not mentioning the 6.03.
Leverage is the part a cash account cannot buy
Bridgewater is explicit that borrowing is what makes the concept work, and the 2012 paper spells the arithmetic out in a footnote. Put $10 in the S&P 500 and $10 in bonds and the stocks dominate the risk. Put $5 in stocks and $15 in 10 year bonds and the risk is balanced, "though with a lower return". Then, in Bridgewater's words, "add a bit of leverage and the portfolio has the same return as the stocks but less risk".
Read that footnote again and the 2.04% stops being a surprise. The unlevered version is only the first half of the recipe. It reaches the balance and then stops, at whatever return that balance happens to pay.
One retail fund does the second half. The RPAR Risk Parity ETF targets 25% of risk in each of TIPS, global equities, commodities and US Treasuries, and reaches the Treasury exposure with futures. Its SEC prospectus of 30 March 2022 gives long-term target allocations of 35% TIPS, 25% global equities, 25% commodities and 15% Treasury bills, with the bills serving as collateral so total notional Treasury exposure runs above 15%. Net of a fee waiver it charged 0.51% then, and the fund's own site shows 0.52% now.
Its record is the counter-evidence to any claim that borrowing fixes this. RPAR returned 7.78% in 2021, lost 22.81% in 2022, then made 6.32%, lost 0.11% and made 18.28%. That compounds to 0.89% a year, 4.51% over five years, and a five-year figure of 1.23% a year on the fund's own reporting to 31 July 2026.
The levered implementation had the deepest 2022 of the four portfolios here: 5.84 points below the 60/40 and 7.75 points below the unlevered stack. Borrowing against long Treasuries in a year when they fell 31.41% did what borrowing does. It made the loss bigger.
The strongest objection: five years is a sample, not a verdict
The serious case against everything above is that 2021 to 2025 was almost tailor-made to embarrass this portfolio, and the sample is tiny.
It contained a historic bond bear market. TLT compounded at -1.93% a year over the full ten years to 30 June 2026 and -6.66% over five. It also contained an extraordinary US equity run: IVV at 15.47% a year over ten years. A portfolio holding 12.0% equities and 57.4% in government bonds of one kind or another was going to trail that, and the reason has nothing to do with whether the risk balance is sound.
The ten-year figures narrow the window less than you'd hope. Weighting each sleeve's own ten-year annualised return by its stack weight gives 3.68%, against 9.49% for the 60/40, a gap of 5.81 points. That calculation is a weighted average of six separate fund returns, not the realised return of a rebalanced portfolio, so treat it as a sense check rather than a result.
For scale on how short five years is: the UBS Global Investment Returns Yearbook, published on 3 March 2026 with Dimson, Marsh and Staunton, covers 126 years across 35 markets. On that record gold's real US dollar price has risen 5.2-fold since 1900, an annualised real return of 1.3%. Our reading of gold demand in 2025 covers the recent end of that series. Five calendar years cannot settle an argument about a portfolio designed for economic regimes that arrive once a decade.
Costs and domicile move the answer for a UK reader
Weighted by the table above, the six-fund stack costs 0.226% a year. Most of that comes from the two largest sleeves rather than the priciest one: TIP at 0.18% on 42.2% of the money, and EMB at 0.39% on 17.8%. The 60/40 costs 0.078%. RPAR costs 0.52%. The fee difference between the stack and the 60/40 is 0.148 points a year, roughly a fortieth of the 6.03-point return gap, so cost isn't the story here.
Domicile matters more. Every fund in the table is US-listed, on NYSE Arca or NASDAQ. The FCA's rules require that "a person who advises a retail investor on a PRIIP or sells a PRIIP to a retail investor must provide the retail investor with a KID in good time before any transaction is concluded", and the PRIIPs regime has now been replaced by the Consumer Composite Investments Regulations. Whether any given US-listed fund reaches a UK retail account turns on that disclosure requirement, not on anything in the arithmetic above. A sterling reconstruction would use different funds, with different fees and different tax treatment, and none of these figures transfers to it untested.
What this arithmetic cannot tell you
Three limits, and the first is the biggest. The box-to-asset split is a reconstruction. Bridgewater publishes 25% per environment and the asset names, and everything downstream of that, the even split inside each box, the choice of one fund per sleeve, the use of standalone volatility, is a decision someone had to make. A different reasonable reading gives different weights and a different result.
Second, standalone volatility ignores correlation. Summing each sleeve's own risk gives 8.36% for the stack and 10.45% for the 60/40, and both overstate what a diversified portfolio actually experienced, because assets that move against each other cancel. The ranking survives; the levels don't. The full covariance version belongs in a different piece.
Third, the sample. Five calendar years of one country's funds is not evidence about a strategy that is supposed to be indifferent to regimes. It is one path. It contained a year when long Treasuries fell 31.41% and a year when US equities rose 28.66%, and both cut the same way.
What the numbers do support is narrower and still useful. Over this window, the unlevered reconstruction achieved the risk balance and delivered a low return, and the levered retail version achieved the balance and delivered a lower one with a deeper drawdown.
What would change the conclusion
A regime where bonds and commodities both work. The five years measured here had one of the four boxes paying and two of them losing badly. A stagflationary decade with a normal bond market would flip the comparison, because 57.4% of this stack sits in government bonds and 6.0% in commodity index returns that have their own roll and collateral mechanics.
Cheap, taxable-account leverage. Bridgewater's footnote says borrowing converts the balanced portfolio's lower return back into an equity-like one. RPAR shows what happens when the borrowing sits against long duration in a rate shock. A retail vehicle that balanced risk without concentrating the borrowing in one asset class would be a genuinely different test, and none of the numbers here would apply to it.
A published benchmark. If Bridgewater ever released an All Weather return series, the phrase "tracking gap" would mean what it usually means, and every reconstruction including this one could be scored against it. Until then, the honest comparison is a reconstruction against a 60/40, which is what a reader is choosing between anyway.
The number worth watching isn't the return. It's the risk share: what fraction of your portfolio's movement one asset class is responsible for. LedgerTouch reports that continuously. The equity risk share of a plain 60/40, 75.0% on these figures, is the observation Bridgewater's own origin story turns on.