Key takeaways
- Across 16 advanced economies, the Jorda-Knoll-Kuvshinov-Schularick-Taylor study found average real total returns of 7.05% a year on housing against 6.89% on equities — with a standard deviation of 9.98% for housing against 21.94% for equities.
- Only 39.94% of US housing's long-run nominal total return came from price. The rest was rental income, which an owner-occupier receives as shelter rather than as cash to reinvest. The US capital gain on its own averaged 3.54% a year in nominal terms.
- The same study's country table has US equities ahead of US housing in all three windows it reports — 8.39% against 6.03% over the full sample, 8.75% against 5.62% post-1950 and 9.09% against 5.66% post-1980, in real terms. The UK ordering is the same.
- The 9.98% figure belongs to a national index. Real US house prices fell 35.9% from peak to trough between 2006 and 2012 on a national basis, but 60.8% in Phoenix and 56.4% in Miami.
- For the 50th-to-90th percentile of US wealth, owner-occupied real estate was 37.6% of assets and 44.0% of net worth at the first quarter of 2026, against 10.6% of assets held in corporate equities and mutual fund shares.
The house is on your balance sheet; the housing returns you have read about are not the returns you receive
Your house is most of your net worth, and you want to know whether that leaves you diversified or dangerously concentrated. The honest answer has two halves, and they point in opposite directions.
Yes, it belongs in the accounting. Leaving out the largest thing you own doesn't make the risk go away; it just stops you seeing it. The Federal Reserve's Distributional Financial Accounts put owner-occupied real estate at 37.6% of total assets for households between the 50th and 90th percentiles of US wealth at the first quarter of 2026, and 44.0% of their net worth. Corporate equities and mutual fund shares came to 10.6% of assets for the same group. A stated allocation that covers only the second number is describing a corner of the balance sheet.
And no, the published housing returns are not yours. The best long-run dataset on the subject reports total returns on the national housing stock, unlevered, with net rent counted as income. An owner-occupier holds one house, usually with a mortgage, and spends the rent by living in it. Each of those three differences is large, and they don't cancel.
The strongest counter-argument: 16 countries, 145 years, and a better Sharpe ratio in every one
The serious version of the argument comes from The Rate of Return on Everything, 1870-2015, by Oscar Jorda, Katharina Knoll, Dmitry Kuvshinov, Moritz Schularick and Alan Taylor — a dataset of bills, bonds, equities and housing across 16 advanced economies. The figures here are from the December 2017 San Francisco Fed working-paper version; the dataset's own documentation cites the study as forthcoming in the Quarterly Journal of Economics.
Their headline global table, equally weighted across the 16 countries, gives an arithmetic mean real total return of 7.05% a year on housing against 6.89% on equities. Housing's standard deviation was 9.98%; equities' was 21.94%. On geometric means the gap widens: 6.61% for housing, 4.64% for equities. After 1950 the ordering flips on returns — equities 8.28%, housing 7.44% — but the volatility gap holds, at 24.20% against 8.88%.
Risk-adjusted, it's not close. The authors write: "Housing provides a higher return per unit of risk in each of the 16 countries in our sample, with Sharpe ratios on average more than double those of equities." Over the long run housing outperformed equities outright in 6 of the 16 countries, equities won in 5, and the two were about level in the remaining 5.
Two things make this stronger than it first looks. The rental yields are already net of maintenance, ground rent, other irrecoverable expenditure and depreciation, so it isn't a gross figure flattered by ignoring the boiler. And the return in kind is real money: the US national accounts put the imputed rental value of owner-occupied non-farm housing at $2,500.3 billion in 2025. That is what American homeowners consumed by not paying a landlord.
Just over 60% of US housing's long-run nominal return was rent, and an owner-occupier eats it
Now the decomposition, which is where the argument turns.
The same study splits total returns into capital gain and income. For the United States, over its housing sample of 1891 to 2015, the nominal figures are a capital gain of 3.54% a year, rental income of 5.33%, a total of 8.87%, and a capital-gain share of 39.94%. So just over 60% of the return came from rent. For the UK, across the 108 annual observations its housing series carries, the split is 5.44% capital gain and 3.94% rent, a capital-gain share of 58.01%.
An investor who owns a rental property banks that income and can redeploy it. An owner-occupier receives it as accommodation. It's genuinely valuable — it's why owning isn't the same as burning money on rent — but it doesn't compound inside the asset, and it can't be rebalanced into anything else. Roughly speaking, the US owner-occupier's house delivers the 3.54% nominal price leg, plus a stream of shelter that arrives as consumption rather than capital.
This is also why house price indices mislead when read as investment returns. Every one of them — Case-Shiller, the UK House Price Index, the BIS series — is price only. The rental yield, the largest single component of US housing's historical return, is absent by construction. So is every cost of ownership.
In the United States and the United Kingdom specifically, equities won every window
The cross-country average hides the country you actually live in. The country table doesn't.
In real terms, US equities returned 8.39% a year against 6.03% for US housing over the full sample; 8.75% against 5.62% after 1950; and 9.09% against 5.66% after 1980. UK equities returned 7.20% against 5.36% for UK housing over the full sample; 9.22% against 6.57% after 1950; 9.34% against 6.81% after 1980. In each case the comparison uses a matched sample within the country, and equities are ahead in every window reported.
The 16-country average that puts housing first is carried by places like France, Belgium, Norway and Denmark. After the Second World War, housing was the best-performing asset class in 3 countries and equities in 9. Which fact applies to you depends on where your house is.
Aggregate volatility is not your volatility: one house, one postcode
The 9.98% standard deviation is the property of a national index, in the same way that an index fund's volatility isn't the volatility of one stock. The authors say so directly: "For example, we found that in the U.S., local (ZIP5) housing return volatility is about twice as large as aggregate volatility, which would about equalize risk-adjusted returns to equity and housing if investors owned one undiversified house."
You can see the same thing in the drawdown data. Deflating the S&P Cotality Case-Shiller national index by US consumer prices and tracking the running maximum, real US house prices fell 35.9% from a May 2006 peak to a February 2012 trough, and didn't regain that level until March 2021. That series is a construction of mine, so here's the check: Robert Shiller's own published real home price index gives 35.7% over the same episode, with the same trough month and the same recovery month, and the BIS's real residential property price index for the United States gives 39.0%. The three sit within 3.3 percentage points of each other.
The metro figures, on the same construction, are what the chart plots: 60.8% in Phoenix, 56.4% in Miami, 55.6% in Detroit, 49.9% in San Francisco, 48.5% in Los Angeles, 46.2% in Chicago and 36.0% in New York. In April 2026, real Chicago house prices were still 16.6% below their October 2006 peak — 19 and a half years later. The national number is an average of experiences almost nobody had.
The mortgage is the position, and it never appears in the return series
The published housing returns are unlevered. Most owner-occupiers are not. The National Mortgage Database counts 51.4 million outstanding residential mortgages at the first quarter of 2026, against 87.3 million owner-occupied housing units in the Census Bureau's count for the same quarter. That's 58.9% at most — at most, because the mortgage count also includes liens on property that isn't owner-occupied. So roughly four in ten American owner-occupiers own outright. The study's authors claim only that "the majority of households in advanced economies today hold a leveraged portfolio in their local real estate market".
The study's authors run the adjustment themselves, using economy-wide leverage of 20% of housing value and a 2.5% real borrowing cost: "In other words, for the representative agent, the levered housing return is about 110 bps higher than the unlevered housing return (8.1% versus 7%), which is a small difference and still leaves equity and housing returns roughly comparable." That is the aggregate. A household's own leverage isn't, and it's highest exactly when the position is newest and least diversified.
The current US picture is milder than the 2008 memory suggests. The Federal Housing Finance Agency's National Mortgage Database records 51.4 million outstanding residential mortgages at the first quarter of 2026, with an average mark-to-market loan-to-value ratio of 45.1%; 74.3% of loans sat at 60% or below and 0.2% were above 100%. In the first quarter of 2013 the same series showed an average of 69% and 11.3% above 100%. Price appreciation and amortisation have de-levered the stock.
That describes the average outstanding loan, not a recent buyer. The 30-year fixed mortgage average was 6.58% in the week to 23 July 2026, and 4.3% of outstanding loans still sat between 91% and 100% of current value. Leverage cuts both ways by arithmetic, and the asset here is a single undiversified building.
Your employer and your house usually share a postcode
This is the risk that gets least attention and deserves the most, because it is a correlation rather than a level.
Phoenix is the clean case. Real house prices in the metro fell 60.8% from June 2006 to September 2011 on the construction above. Total non-farm employment in the Phoenix-Mesa-Chandler metro area peaked at 1,926,200 in October 2007, fell to 1,683,000 by September 2010 — a drop of 12.6% — and didn't regain its 2007 level until September 2015. The house and the job market broke in the same place, at roughly the same time, and they didn't break independently.
For a household whose wealth is a local house and whose income is a local job, those aren't two independent exposures. They are one bet on one regional economy, made twice, and the second leg is the one that pays the mortgage. A globally diversified equity portfolio has the opposite property: its worst years are rarely correlated with a single metro labour market.
Every published house price index leaves out what ownership costs
Transaction costs are the part the charts silently drop. Citing OECD data, the study puts average round-trip transaction costs at about 7.7% of a property's value across 13 of its 16 countries, from nearly 15% in Belgium to about 1% in Denmark. Annualised, the authors conclude: "Accounting for transaction costs would thus lower the average annual return to housing by less than 100 basis points (e.g., 77 basis points per year based on a 7.7% cost incurred every 10 years)." Equity transaction costs, annualised over observed turnover, have historically been of a similar order — so this is a real cost, not a decisive one.
Taxes are excluded from those return figures entirely, and they are not small. In England and Northern Ireland, stamp duty land tax on a single residential property is zero up to £125,000, 2% on the portion to £250,000, 5% to £925,000, 10% to £1.5 million and 12% above that, with usually 5% on top if the purchase means owning more than one. On the UK average price of £270,080 in April 2026, those bands work out at £3,504 before any first-time-buyer relief — about 1.3% of the price, payable on the way in.
Then there is time. The median US home listing spent 53 days on the market in June 2026, in a series that has ranged from 30 days to 88 days since 2016. Most owners aren't trading anyway: more than half of US homeowners in the 2007 American Community Survey had lived in their current home for more than 10 years.
An asset you cannot rebalance changes what a 60/40 means
Rebalancing is the mechanism that makes a stated allocation mean anything. It's also the thing a house can't do. You can't sell 7% of a kitchen because equities ran.
Once the house is folded in, the arithmetic moves. Owner-occupied real estate is 37.6% of assets for the 50th-to-90th percentile group in the Distributional Financial Accounts, while corporate equities and mutual fund shares are 10.6%. Whatever the pension statement says the equity weight is, the whole-balance-sheet weight is a fraction of it, and the largest single line is an illiquid, levered, geographically concentrated real asset that drifts wherever the local market takes it. The only lever that moves the total is the financial sleeve. How often that sleeve is rebalanced is a smaller question than what it is being rebalanced around.
US homeownership was 65.3% in the first quarter of 2026, within a range running from 62.9% in 1965 to 69.2% in 2004. This is the modal household position, not an eccentric one. Tools that report a portfolio — LedgerTouch included — mostly report the tradeable part, which is the easiest part to measure and the smaller part of the picture.
What the evidence cannot settle
The 16-country dataset stops in 2015. It's history, assembled from indices with real measurement error, and the housing samples begin at different dates in different countries — 1891 for US housing, 1900 for the UK, 1948 for Portugal. Comparisons across countries aren't comparisons across identical periods.
The rent-price ratios that anchor the housing returns are benchmarked and then extended with price and rent indices. That's a modelling choice, and the authors are explicit that running costs are assumed stable relative to gross rent within each country.
My real house price series are constructions, not published indices: nominal index divided by consumer prices, expressed in April 2026 money. For the United States that construction sits between two published real series, at 35.9% against Shiller's 35.7% and the BIS's 39.0%. For the United Kingdom, my method gives a 29.3% real fall from July 1989 to December 1995 with recovery in January 2000, against 28.9% from 1989-Q3 to 1996-Q2 with recovery in 2000-Q1 on the BIS index — a validation error of 0.4 percentage points. That same UK episode was only a 12.3% fall in nominal terms. On the nominal chart alone, that crash barely happened.
Nothing here knows your tax code, your pension rules, whether your house is in a supply-constrained city or a shrinking one, or whether your income depends on the same local economy your house sits in. National averages are the wrong unit for a question about one building.
What would change the conclusion
If the rental leg were smaller than the study finds. The whole argument rests on the decomposition: 60% of US housing's return arriving as rent is what separates the investor's housing return from the owner-occupier's. The authors name that as the soft spot themselves — the biases specific to their housing estimates, they write, "include the costs of running a housing investment, and the benchmarking of rent-price ratios to construct the historical rental yield series." The authors have already run part of that test. Their Table 9 rebuilds the global housing return on each country's lowest and highest available rental-yield benchmark: 6.26% at the low against equity's 6.89%, 7.89% at the high. Their reading is that "for all benchmark values, returns on housing are similar to those on equity". It would take a benchmark outside that range. If US net rental yields were materially lower than 5.33%, the case for housing as an asset class and the gap between owning and letting out would narrow together.
If you own more than one property. Everything above turns on the rent being consumed. A second property that is let out receives the income in cash, can be sold in slices across a portfolio of properties, and behaves much more like the 7.05% asset in the table. The owner-occupied case is the special one.
If your local market decouples from your income. The Phoenix pattern — house and payrolls falling together — is the reason concentration matters more than the volatility number suggests. A household whose income is genuinely portable, or whose house sits in a metro uncorrelated with its employer, carries a materially different risk from the one described here.
If you are outside the US and the UK. In France, Belgium, Norway and Denmark, housing beat equities over the long run, and the ordering that makes this piece's argument sharpest in America reverses. The same goes for transaction costs, which run from about 1% to about 15% depending on the border.
None of this resolves into a target weight, because there's no rebalancing trade at the end of it. What it resolves into is a fraction: your house, at market value, divided by everything you own. For most households that number is closer to the 44% of net worth in the Federal Reserve's data than to zero, and the question it raises is what the other half is doing — a question that looks different once the house is on the page instead of assumed away, and one that also governs how any other single-asset sleeve gets sized.