A 60/40 portfolio isn't two bets. It's one bet with a cushion. Using the long-run volatilities in the Jordà–Schularick–Taylor return database — 21.8% a year for equities, 10.7% for bonds — equities supply roughly 81% of total portfolio risk when the two correlate at plus 0.5. Push that correlation to minus 0.3 and equities supply 99% of the risk, because the bonds spend their time cancelling equity moves rather than adding wobble of their own.
So the case for extending to five sleeves doesn't rest on owning five different things. It rests on whether the extra three change where the risk sits, and what they charge for it.
The maths that decides it
Portfolio variance is the weighted sum of each asset's variance, plus twice the weighted sum of every covariance. For two assets that's w1²σ1² + w2²σ2² + 2w1w2ρσ1σ2. The third term is the whole argument. When ρ is small, the cross term shrinks and portfolio volatility drops below the weighted average of the parts.
Run it on the 60/40 above. At a correlation of minus 0.3, portfolio volatility works out at 12.5%. At plus 0.5 it's 15.7%. That's 26% more risk from identical weights and identical assets. The World Gold Council ran a comparable exercise on US equities, Treasuries and gold using monthly data from January 1973 to April 2025, and found a 17% rise in portfolio risk when the bond–equity correlation flipped from negative to positive. The exact figure depends on the assets and weights. The direction doesn't.
Marginal contribution to risk is the second half of the maths, and it's the half most people skip. An asset's risk contribution is its weight times its covariance with the whole portfolio, divided by portfolio volatility. Contribution scales with weight and with correlation — not with the sleeve's own headline volatility. That single fact explains most of what follows.
Take bitcoin. Its annualised realised volatility was 52.2% at the end of the first quarter of 2025, against 15.5% for gold, on NYDIG's figures. That sounds like a portfolio-wrecking asset. At a 2% weight it isn't. Add 2% bitcoin to the 60/40 and modelled volatility moves from 15.7% to 15.4% if bitcoin is uncorrelated with equities, or to 15.9% if that correlation is 0.6. Bitcoin's own share of portfolio risk lands somewhere between 0.5% and 3.6%. A 5% gold sleeve at zero correlation works in the other direction: volatility falls to 14.9%, and gold accounts for 0.3% of the risk.
Now the full five-sleeve version. Take 45% equities, 30% bonds, 5% gold, 18% property and 2% crypto. Assume a bond–equity correlation of 0.5, gold uncorrelated with everything, property correlated at 0.1 with both stocks and bonds, and crypto at 0.3 with equities. Modelled volatility falls from 15.7% to 12.4%. But equities still supply 76% of the risk, bonds 17%, property 3.6%, crypto 2.7% and gold 0.4%. Five sleeves, one dominant risk factor. The extra sleeves changed the size of the risk, not its source.
What each sleeve added, and what it charged
Equities did the compounding. One US dollar invested in US equities in 1900 grew to 124,854 dollars in nominal terms by the end of 2025, against 284 dollars for long bonds, 69 dollars for Treasury bills and 38 dollars for the inflation index. Those are the figures in the UBS Global Investment Returns Yearbook 2026, built on the Dimson, Marsh and Staunton database. Equities were the best-performing asset class in all 21 countries with continuous investment histories.
Bonds bought drawdown relief rather than return. Since 1900, equities and bonds have each lost more than 70% in real terms on several occasions. A 60:40 blend of the two has never fallen more than 50%. That's the covariance term showing up as lived experience, and it's the reason the two-asset portfolio became the default in the first place.
Gold's record splits in two. The real dollar gold price has risen 5.2-fold since 1900, an annualised real return of 1.3%. Over the 54 years since the end of Bretton Woods it returned 4.7% a year in real terms for a US investor, 5.8% in the UK and 4.3% in Switzerland. Its inflation-hedging reputation is shakier than the folklore suggests: of the 28 years in which inflation exceeded 3%, gold's return was negative in 13 of them.
The recent portfolio evidence is kinder. FTSE Russell compared a 60/20/20 equity–bond–gold mix against a straight 60/40 from 2010 to March 2025. The gold version returned 7.5% a year against 6.3%, with volatility of 8.55% against 8.01%, lifting the Sharpe ratio from 0.25 to 0.38. Fifteen years is a short window, and it happens to cover the stretch when the bond–equity correlation turned positive. The longer record behind the 5-10% gold allocation case and its 45-year drawdown is considerably less flattering.
Property looks best of all on the index numbers, and that's exactly the problem. Across 16 advanced economies from 1870 to 2015, Jordà and co-authors put real housing returns at 7.06% a year against 6.88% for equities, with a standard deviation of 9.93% against 21.79%. Compounded, housing returned 6.62% a year to equities' 4.66%. Same arithmetic return, half the volatility, better compounding. If that were investable as stated, the asset-allocation debate would already be over.
Property isn't held the way the backtest assumes
It isn't investable as stated. Three things break the translation.
Costs come first. The same paper cites OECD estimates that round-trip transaction costs average about 7.7% of a property's value across 13 of the 16 countries studied. On a ten-year holding period that's roughly 77 basis points a year off the return, before letting agents, maintenance, insurance and tax. Nothing in the equity sleeve costs anything close to that.
Second, the index isn't a house. A national house price series averages away everything specific to one building, one street and one tenant, and nobody owns the average. The 9.93% standard deviation is the volatility of a country's housing market, not of an asset anyone actually holds. Appraisal-based and repeat-sale indices also update slowly, which suppresses measured volatility further and flatters every correlation calculated from them.
Third, concentration and leverage. Net property wealth made up 40% of household wealth in Great Britain between April 2020 and March 2022, ahead of private pensions at 35%, with net financial wealth at 14% and physical wealth at 10%. Median household wealth was £293,700. One caveat on that source: the Office for Statistics Regulation suspended the accredited-official-statistics badge for this wealth survey in June 2025, citing quality and response-rate concerns, so treat the shares as indicative of the shape rather than precise. For most households the property sleeve isn't 18% of a portfolio. It is the portfolio, bought with a mortgage, and it can't be rebalanced in 5% slices. The gap between index housing and owned housing is the subject of our piece on treating your house as a portfolio asset against equities.
Crypto isn't held that way either
The crypto sleeve has the opposite problem. It's usually far too small to matter. FCA survey work found 12% of UK adults owned cryptoassets in August 2024, around 7 million people, up from roughly 10% in 2022 and 4.4% in 2021. Among owners, 32% held £100 or less, and only 4% held more than £10,000. In the 2025 wave, 73% bought through a centralised exchange and 91% of the public had heard of cryptoassets.
A £100 position is not a 2% sleeve of anything except a very small portfolio. The people who genuinely run crypto at a policy weight are a thin slice of a thin slice. Everyone else holds a lottery ticket that the diversification maths has no opinion about.
There's a rebalancing problem too. A backtest assumes annual or threshold rebalancing, which means systematically selling crypto after it triples and buying it after it halves. That's the mechanism producing most of the modelled benefit. Whether holders on centralised exchanges behave that way is an empirical question, and the survey evidence doesn't answer it. Our look at what CoinShares crypto fund flow data really measures covers how little the flow numbers reveal about holding behaviour.
The case against extra sleeves
The strongest objection has nothing to do with correlations. Critics argue that extra sleeves add cost, tax friction, tracking error and — most expensively — more chances to behave badly. Morningstar's Mind the Gap 2025 study puts a number on that last one.
Over the ten years to 31 December 2024, the average dollar in US funds earned 7.0% a year against an 8.2% total return. The gap of 1.2 percentage points is worth about 15% of the funds' gains. It wasn't spread evenly. Allocation funds — the one-decision, multi-asset kind — gave up only 0.1 points. US equity funds gave up 0.6. Sector equity funds, the most specialised category, gave up 1.5. Morningstar also found that the further a fund strayed from its benchmark, and the more its investors traded, the wider the gap grew. That headline figure is contested: Fulkerson, Jordan, Riley and Yan, writing in the Financial Analysts Journal in 2026, rebuilt the calculation on the same sample and put the cost of poor timing at 0.10% a year rather than 1.2%.
That lands squarely on the five-sleeve argument. Specialised, volatile, narrative-heavy sleeves are precisely the ones people buy after a run and sell after a fall. A gold sleeve that theoretically trims 5% off portfolio volatility can easily cost more than that in mistimed entries. The mechanics of that leakage are covered in our piece on what the behaviour gap costs investors in timing.
The evidence isn't one-sided, though, and honesty demands the awkward bit. Alternative funds showed a gap of just 0.1 points over the same decade. Morningstar notes it would have been 0.8 points had liquidated funds been excluded, since those funds are treated as earning no return after they close. Measurement choices move the answer by a factor of eight, which is a useful reminder about how much weight any single behaviour-gap number can carry.
What would change the conclusion
Four things, in rough order of likelihood.
First, the correlation regime. The World Gold Council's data shows the US equity–Treasury correlation has generally been negative when core inflation ran below 2.5%, and has broken down above it. If inflation settles under that threshold, bonds resume the diversifying job for nothing, and the marginal case for gold and crypto weakens sharply. Most of the recent enthusiasm for extra sleeves is an artefact of one correlation flipping sign.
Second, crypto's correlation with equities. The 2% sleeve above contributes 0.5% of portfolio risk at zero correlation and 3.6% at 0.6. If it keeps drifting towards equities, it stops being a diversifier and becomes leveraged equity in a costlier wrapper. That's a testable claim, and it will resolve itself within a few years.
Third, how property risk is measured. The 12.4% five-sleeve figure leans on the 9.93% index standard deviation for housing. Individual properties are lumpier, less liquid and far more idiosyncratic than any index, and the true number for an owned asset is higher. How much higher is genuinely unsettled. If it's a lot higher, the property sleeve's contribution shrinks and the modelled benefit goes with it.
Fourth, concentration inside the equity sleeve. The UBS Yearbook notes that by the end of 2025, US equity market concentration was at its highest level for at least 100 years, and that correlations between developed and emerging markets, and between equities and bonds, have both risen. If the equity sleeve is itself becoming one bet, the argument for holding something genuinely unrelated gets stronger rather than weaker.
What the arithmetic supports is narrower than either camp tends to claim. Extending from two sleeves to five cut modelled volatility from 15.7% to 12.4% — a real reduction, not a rounding error. It did not change the fact that equities supplied three-quarters of the remaining risk. And the modelled gain assumes costs of zero, perfect rebalancing and stable correlations, none of which survive contact with an actual property or an actual exchange account. The record on which asset classes actually diversified during a crisis is where that gap between model and outcome shows up most clearly.