Risk Parity: The Financing Spread Ate Half the Advantage

Key takeaways

• In the 1926 to 2010 study that made the academic case, risk parity beat the US market portfolio by 4.15 percentage points a year when financed at Treasury bill rates.
• Financing the same portfolio at LIBOR, which averaged 62.3 basis points over bills, cut that to 1.81 points. Sixty-two basis points took 56% of the advantage.
• Interactive Brokers charges a UK investor 5.227% on the first 80,000 pounds of a sterling margin loan, a published spread of 1.5 percentage points over its own benchmark.
• That portfolio ran about 3.55 times leverage. Borrowing 2.55 pounds per pound of equity at a 1.5 point spread costs 3.82 points a year, or 92% of the 4.15 point edge.
• Measured in real terms across 20 countries from 1900 to 2025, equities were 1.57 times as volatile as bonds, not 5.8 times. That version needs 1.18 times leverage, not 3.55.

The short answer: for risk parity, the financing spread costs more than the fund fee

You've probably seen the chart. Equal risk from stocks and bonds, borrow to bring the whole thing up to normal portfolio volatility, and the result beats a 60/40 on every risk-adjusted measure. You want to know what it costs to run that at home.

The fee isn't the answer. The two US-listed funds that package the strategy charge 0.50% and 0.65% a year, and one of them turned over just 17% of its book in 2025. Neither number is large enough to matter much.

The cost is the spread on the borrowing. Asness, Frazzini and Pedersen tested their own portfolio at five different financing rates in the 2012 Financial Analysts Journal paper, and the result is the whole story of this article. At Treasury bills, the strategy beat the market portfolio by 4.15 percentage points a year over 1926 to 2010. At one-month LIBOR, an average of 62.3 basis points higher, the same portfolio beat it by 1.81 points. Sixty-two basis points of spread took 2.34 points, which is 56% of the advantage.

A UK retail margin loan does not cost 62 basis points over the benchmark. It costs 150.

Why the strategy has to borrow at all

The idea is simple and the arithmetic behind it is worth two minutes. In a conventional 60/40, almost all the portfolio's movement comes from the equity side, because equities are the volatile part. Equalising the risk contribution means holding far more bonds than stocks.

How much more depends entirely on the measured volatilities. In the AQR sample, US stocks ran 19.05% annualised volatility against 3.28% for US bonds. Inverse volatility weighting, which sizes each holding by the reciprocal of how much it moves, is the construction the paper uses. It gives roughly 15% stocks and 85% bonds. That is what the authors report as the average unlevered allocation across the whole sample.

A portfolio like that is very quiet. Its measured volatility was 4.25% a year, against 11.68% for a 60/40 and 15.08% for the market. Quiet also means low absolute return, so the strategy borrows to scale the whole thing up. The paper multiplies the weights by a constant chosen to match the market portfolio's 15.08% volatility. Dividing 15.08 by 4.25 gives about 3.55, so the portfolio held roughly 355% of its equity in assets and borrowed the other 255%.

That borrowed 255% is where the running cost lives. Every basis point of spread over the risk-free rate gets paid on all of it, every year, whether the strategy works or not. This is a different failure mode from the one a leveraged portfolio faces in a drawdown, and it is more boring: it just leaks.

The founding paper left the financing cost out, and said so plainly

The paper is called Leverage Aversion and Risk Parity, and leverage aversion is its whole argument. If enough investors refuse to borrow, safer assets have to offer higher risk-adjusted returns than riskier ones. Someone willing to borrow can collect that difference. The cost of borrowing is not a footnote to the thesis, then. It is the thesis. And this is not a criticism the authors would resist, because they wrote the omission down themselves. The paper says its simulated levered results do "not reflect any adjustment of the returns for the costs of leverage, such as financing spreads and costs associated with deleveraging." It adds that at modest leverage those costs "should be quite low."

Appendix B is the part almost nobody quotes. It reruns the headline result at the repo rate, the overnight indexed swap rate, the effective federal funds rate and LIBOR, sorted by their average spread over Treasury bills. The advantage over the market portfolio falls from 4.15 points at bills, to 3.38 at repo, to 3.21 at OIS, to 2.64 at fed funds, to 1.81 at LIBOR. The chart above plots that table.

Every rate in that list is an institutional one. The widest, LIBOR at 62.3 basis points over bills, is roughly what a bank paid another bank. It is not what a private investor pays a broker.

A UK margin loan costs 1.5 points over the benchmark, on every borrowed pound

Interactive Brokers publishes its sterling margin loan rates in tiers. As at 28 August 2026, the first 80,000 pounds of a GBP margin loan is charged at 5.227%, which the firm labels as its benchmark plus 1.5%. The next tier, from 80,000 to 800,000 pounds, is benchmark plus 1%. From 800,000 to 38 million pounds it drops to benchmark plus 0.75%, and above 38 million to benchmark plus 0.5%. The rate is blended, so the first slice is charged at the first-tier rate regardless of how large the loan gets.

For scale, the Official Bank Rate has been 3.75% since 18 December 2025. The first-tier charge of 5.227% sits 1.48 points above it, which is close to the 1.5 point spread the schedule advertises.

Now put that back into the arithmetic. At 3.55 times leverage you borrow 2.55 pounds for every pound of your own. A 1.5 point spread on 2.55 pounds costs 3.82 pence a year per pound of equity, so 3.82 percentage points of annual return. The advantage the paper measured, before any financing spread, was 4.15 points. The retail spread takes 92% of it.

That calculation is mine, not the paper's, and it deserves the caveat: it assumes you hold the leverage constant and that the whole balance sits in the first tier. A larger loan blends into cheaper tiers and the drag falls. It never reaches zero, because the cheapest published sterling tier is still 0.5 points over the benchmark, and that one starts at 38 million pounds.

The packaged version costs 0.65% a year and trades less than you would guess

The other route is to buy the strategy rather than build it. Two US-listed funds do this in public. In its annual report for the year ended 31 December 2025, RPAR Risk Parity ETF reported a unitary fee of 0.50%, reduced by contractual waiver to 0.48% through at least 30 April 2026. UPAR Ultra Risk Parity ETF, which runs the same idea with more borrowing, reported 0.65%, waived to 0.63%. The extra leverage costs 0.15 percentage points more in the wrapper.

Two things in RPAR's annual report cut against the usual assumptions about this strategy. The first is turnover: 17% for the year ended 31 December 2025, on net assets of 562.7 million dollars across 104 holdings. Whatever else risk parity costs, it is not a high-frequency trading bill. Anyone budgeting for the strategy on the assumption of constant rebalancing is budgeting for the wrong thing.

The second is how the leverage is actually obtained. At the same date RPAR held 869 US Treasury 10 Year Note futures with a notional value of 97,708,188 dollars, and 814 Ultra US Treasury Bond futures at 96,052,000 dollars. That is 193.8 million dollars of notional against 562.7 million of net assets, about 34%. UPAR held 46.2 million of futures notional against 63.6 million of net assets on the same date, or 72.7%, which is what the extra leverage looks like on a balance sheet. The fund borrows through the futures market, not through a margin loan, and the paper is explicit that futures returns are already excess returns, so "no assumption of the financing cost is needed."

RPAR returned 18.28% in 2025 against 18.83% for its index, a gap of 0.55 points on a 0.48% net fee. UPAR returned 24.38%. These are one year of a US-listed fund, not a UK guide to what your account did.

You cannot pledge an ISA, so the borrowed version sits in a taxable account

Here is the constraint that has no equivalent in any of the American research. HMRC's guidance to ISA managers requires the terms and conditions to specify in writing that "the ISA investments will be, and must remain in, the beneficial ownership of the investor and must not be used as security for a loan". The guidance attaches no exception to that condition.

So a margin-financed portfolio cannot live inside a stocks and shares ISA. It sits in a general investment account, where the gains are taxable. From 6 April 2026 that means 18% on gains inside the basic rate band and 24% above it, after an annual exempt amount of 3,000 pounds for the 2026 to 2027 tax year.

The turnover figure above softens this considerably. A strategy that trades 17% of its book a year realises gains slowly. But the tax is a real layer on top of the financing spread, and it belongs in any honest account of the cost of investing in the UK for this particular strategy.

How much leverage risk parity needs depends on whether you measure in nominal or real terms

The 3.55 times figure is not a property of the strategy. It falls straight out of a bond volatility of 3.28%, and that number is a nominal one measured on US Treasuries.

The 2026 UBS Global Investment Returns Yearbook, built on the Dimson, Marsh and Staunton database, measures the same thing in real terms across 20 foreign countries. From 1900 to 2025 the average standard deviation of real equity returns in local currency was 23.4% a year. For real bond returns it was 14.9%. Equities were 1.57 times as volatile as bonds, not 5.8 times. Two things separate the two measurements. One is the deflator. A conventional bond's coupons are fixed in nominal terms, so an inflation surprise passes straight through to its real return. The other is the sample: 20 countries over 126 years, against US Treasuries over 85.

Run the same inverse-volatility construction on those numbers and the portfolio looks nothing like 15/85. The table below assumes zero correlation between the two, because the summary edition publishes standard deviations and not the covariance.

The middle and bottom rows are my own arithmetic on the yearbook's published standard deviations. As a check on the method, running it on the AQR volatilities returns a 60/40 volatility of 11.5% against the 11.68% the paper reports, and a risk parity volatility of 4.0% against 4.25%, so the zero-correlation shortcut is close enough to trust for a comparison.

The cost implication is large. At 1.18 times leverage you borrow 0.18 pounds per pound, and a 1.5 point spread costs 0.27 percentage points a year. At 1.37 times it costs 0.56. Both are survivable numbers. The 3.82 point version is not, and the difference between them is a measurement choice about which volatility you feed the model, not a difference in the strategy.

The strongest case against all of this

Three objections deserve a hearing, and the first is the serious one.

The futures point. Nobody who runs this strategy properly borrows on margin, and the paper says as much: futures leverage requires no financing assumption because the return series is already an excess return. RPAR's own book shows it in practice. If you can access Treasury futures at institutional roll costs, the 3.82 point number does not describe you. It describes an investor doing it the naive way, which is also the way most private investors have access to.

The second is that the edge survived even the worst financing rate tested. At LIBOR the strategy still beat the market by 1.81 points a year over 85 years, and its Sharpe ratio in the long sample was 0.53 against 0.40 for a 60/40. Halving an advantage is not the same as destroying it. In the United Kingdom specifically, over 1986 to 2010, risk parity returned an average excess 3.14% against 2.62% for a 60/40.

The third is that the leverage requirement itself is contestable, which cuts both ways. If the honest number for a global investor is nearer 1.18 times than 3.55, then the financing spread is a rounding error and this whole article is about a problem that only exists for people who calibrate on post-war US nominal bond data. That is a real possibility, and it is the version of the strategy that the unlevered all weather portfolio approximates without borrowing anything.

What these numbers cannot tell you

The 4.15 point advantage is a backtest over a single 85-year run of US history, and the authors say so. It is one draw. The 1900 to 2025 yearbook figures are cross-country averages of standard deviations, which flatten the variation between a Swiss bond market and an Italian one.

The zero-correlation assumption in the table is a simplification. A positive stock-bond correlation raises the measured volatility of both portfolios, and it raises the equity-heavy 60/40 more, so the leverage multiples in the last column are probably a little high rather than a little low.

Margin rates are a snapshot. The Interactive Brokers schedule is as published on 28 August 2026, tiers move without notice, and the benchmark moves with the Bank Rate. And volatility measured as a standard deviation is not the risk anyone experiences in a bad year, which is the whole argument in volatility vs drawdown. The yearbook notes that since 1900 both equities and bonds have lost more than 70% in real terms on several occasions, while a 60:40 blend has never fallen more than 50%.

What would change the conclusion

Two things would, and neither is a market forecast.

The first is access. If a retail platform offered sterling financing at a spread comparable to the 62.3 basis points the paper's worst case tested, the arithmetic in this piece stops being interesting. The cheapest published sterling tier at Interactive Brokers is 0.5 points over benchmark, and it starts at 38 million pounds of borrowing. Watch the gap between the first tier and the last, not the headline rate.

The second is the bond volatility input. Everything above turns on it. If real bond volatility settles back toward the 12.3% of the 1972 to 2025 window rather than the 14.9% of the full record, the required leverage rises and the financing spread matters more. If it stays where the last few years put it, the strategy needs less borrowing than its own literature assumes, and the cost question mostly goes away. That single input, not the fee and not the trading, is what decides whether risk parity is expensive to run.