The Efficient Frontier Explained: 98 Years of Returns and What Hedge Fund Managers Should Take From It

The efficient frontier is the set of portfolios that deliver the highest expected return for each level of risk, a concept Harry Markowitz set out in the Journal of Finance in 1952. Rebuilt on annual US data for 1928 to 2025, the lowest-risk mix of equities, 10-year Treasury bonds and gold is about 13 per cent equities, 76 per cent bonds and 11 per cent gold, at 6.90 per cent volatility. That is below the volatility of every asset on its own. At the 7.90 per cent volatility of an all-bond portfolio, the frontier earned 7.35 per cent a year against 4.82 per cent for bonds. The arithmetic is sound, but the inputs are less stable than a single 98-year average suggests, and hedge fund return series violate them further. This article rebuilds the frontier, tests where it breaks, and sets out what that means for a manager whose fund must earn a place in an allocator's portfolio.

Every allocator we speak to runs some version of this calculation, even if they would never call it an efficient frontier. The question they are asking is simple: does this fund improve the portfolio we already own, after fees, at the liquidity on offer. Managers who understand that arithmetic present correlation, drawdown behaviour and dealing terms with the same care as returns. The ones who struggle present a backtest with a frontier chart attached and assume the optimiser has made the argument for them.

David Lloyd, Chief Executive Officer at CV5 Capital

Executive Summary

Mean-variance optimisation is the common language of portfolio construction. Allocators use it, consultants build asset allocations with it, and many quantitative and multi-asset strategies run an optimiser inside the fund. Its outputs are only as reliable as three estimated inputs, and the 98-year US record shows how much those inputs move.

• Portfolio return is a weighted average of asset returns. Portfolio risk is not: it is the square root of w transpose C w, which is why combining weakly correlated assets lowers risk.
• On 1928 to 2025 data, the minimum-variance portfolio of equities, bonds and gold carried 6.90 per cent volatility, below bonds at 7.90 per cent.
• The inputs are regime dependent. The stock and bond correlation ran at minus 0.65 from 2000 to 2021, and both assets fell about 18 per cent in 2022.
• Gold traded at a fixed price under the gold standard before 1971, so a single 98-year gold series joins two different assets.
• Reported hedge fund returns are often smoothed, which understates volatility, tends to understate correlation and flatters the fund's position on any frontier.
• Performance of an optimised or backtested portfolio, shown in an advertisement by an SEC-registered adviser, is hypothetical performance under the marketing rule, with specific conditions attached.

How the Efficient Frontier Is Built

Three inputs drive every efficient frontier. The first is mu, the vector of expected returns for each asset. The second is C, the covariance matrix, which holds each asset's variance on the diagonal and each pair's covariance off it. The third is the set of constraints the portfolio must respect.

A portfolio's expected return is the weighted average of the asset returns, written w · mu. Its variance is w transpose C w. Volatility, the usual measure of risk, is the square root of that variance. The difference between the two formulas is the whole of diversification. Return scales linearly with the weights. Variance scales with the squares of the weights, plus cross terms that depend on correlation. Where correlation is low, the cross terms add little and the squared weights shrink each asset's contribution.

The frontier is traced by solving the same problem repeatedly. For each target return, find the weights that minimise variance, subject to the weights summing to one, the return hitting the target and, in a long-only portfolio, no negative weights. The lowest point on the resulting curve is the global minimum-variance portfolio. Every portfolio on the curve above it is efficient. Every portfolio below it, or to its right, is dominated: another mix offers more return for the same risk or less risk for the same return.

Rebuilding the 98-Year Frontier

The calculation below uses the annual return series for 1928 to 2025 published by Professor Aswath Damodaran at NYU Stern: US large-capitalisation equities with dividends, the 10-year US Treasury bond and gold. CV5 Capital computed every figure independently from the source file.

Figures current as at 14 October 2026, computed by CV5 Capital from the dataset as downloaded on that date. The diversification effect is visible in a simple two-asset case. A portfolio of 80 per cent bonds and 20 per cent equities has variance of 40.0 from bonds, 15.1 from equities and 0.9 from the covariance term, a total of 55.9. That is 7.48 per cent volatility, against 7.90 per cent for bonds alone, while the expected return rises from 4.82 to 6.23 per cent.

Solved exactly as a long-only problem, the global minimum-variance portfolio holds 12.8 per cent equities, 76.0 per cent bonds and 11.2 per cent gold, at 6.90 per cent volatility and a 6.01 per cent expected return. The highest-return portfolio at the bonds' own 7.90 per cent volatility holds 30.4 per cent equities, 54.4 per cent bonds and 15.3 per cent gold, and returned 7.35 per cent, more than half as much again as bonds at the same risk.

Why the Frontier Is Less Stable Than It Looks

A 98-year average looks authoritative. It is a single estimate of a moving quantity. Splitting the same data by period changes every input and every answer.

Figures current as at 14 October 2026. Four problems sit behind the table.

Gold is two different assets

The dataset's own notes state that gold prices were fixed under the gold standard before 1971. Gold's pre-1971 volatility of 8.2 per cent reflects administered prices, not a market. Its post-1971 volatility of 27.1 per cent is the asset an investor would actually hold. Averaging the two produces a gold that never traded. The series is also sensitive to single years: removing 2025, when gold returned 66.2 per cent, cuts its 98-year average from 7.36 to 6.75 per cent.

Correlation changes sign

The full-period stock and bond correlation of 0.02 is an average of regimes that look nothing like it. Measured on the same annual data, it was minus 0.09 from 1928 to 1969, plus 0.33 from 1970 to 1999 and minus 0.65 from 2000 to 2021. In 2022, equities fell 18.0 per cent and the 10-year Treasury fell 17.8 per cent. A frontier built on the 2000 to 2021 correlation assumed a hedge that was absent precisely when it was needed.

Expected returns are estimated with large error

The standard error of the 98-year equity mean is 1.96 percentage points, so a conventional 95 per cent interval spans roughly 8.0 to 15.7 per cent. Optimisers treat the point estimate as known. Resampling the 98 years with replacement 1,000 times, the equity weight in the highest-return portfolio at 7.90 per cent volatility ranged from 17 to 41 per cent between the 5th and 95th percentiles. Richard Michaud described the problem in the Financial Analysts Journal in 1989: mean-variance optimisation tends to maximise the effects of errors in its inputs.

Optimised weights often lose to simple ones out of sample

DeMiguel, Garlappi and Uppal, in the Review of Financial Studies in 2009, tested 14 optimisation models across seven datasets. None consistently beat equal weighting out of sample on Sharpe ratio, certainty-equivalent return or turnover. They estimated that a sample-based mean-variance strategy needs around 3,000 months of data to outperform equal weighting with 25 assets. The practical response is not to abandon the frontier but to constrain it, shrink the return estimates, and test how far the weights move when the inputs move.

Why Hedge Fund Returns Break the Inputs Further

Stocks, bonds and gold are priced daily in deep markets. Hedge fund returns are not, and the differences matter more for an allocator placing a fund on a frontier than for an analyst plotting asset classes. Getmansky, Lo and Makarov showed in the Journal of Financial Economics in 2004 that hedge fund returns are often highly serially correlated. Their most likely explanation is exposure to illiquid securities, whose reported returns are smoother than true economic returns. Smoothing understates volatility and inflates the Sharpe ratio, and it also tends to understate measured correlation and beta with markets.

Digital asset funds add a further layer. Volatility in the underlying markets is high enough that a small allocation can dominate a portfolio's risk budget, and correlations with equities have varied across market cycles. An allocator's frontier will size such a fund by its risk contribution, not by its headline return. Managers can address that directly with disciplined position limits, as discussed in our analysis of volatility targeting in hedge fund risk management.

How Allocators Use the Frontier to Judge a New Fund

An allocator rarely asks whether a fund is efficient in isolation. The question is whether adding it moves the existing portfolio towards a better frontier. Mean-variance algebra gives a precise answer. A small allocation to a new fund raises the Sharpe ratio of an existing portfolio only if the fund's Sharpe ratio exceeds its correlation with that portfolio multiplied by the portfolio's own Sharpe ratio. A fund with a modest Sharpe ratio and low correlation can therefore earn a place that a fund with a higher Sharpe ratio and high correlation cannot.

That condition explains several features of institutional due diligence. Allocators ask for monthly returns from inception to estimate correlation and drawdown behaviour, not only annual figures. They test returns against the factors they already own. They read the dealing terms because a fund that cannot be rebalanced cannot sit on a frontier that assumes rebalancing. The questions they ask, and how managers should prepare, are covered in our guide to the hedge fund due diligence questionnaire.

For a launching manager, three practical points follow.

• Report returns at a frequency and with a valuation policy that let an allocator estimate risk honestly. Smoothed or infrequent marks invite an unsmoothing adjustment that will usually raise the fund's measured risk.
• Show correlation and drawdown against the allocator's likely core holdings, and show them in stress periods such as 2008 and 2022, not only on full-period averages.
• Align liquidity terms with the underlying portfolio. Terms that are looser than the assets support create the redemption risk examined in our hedge fund liquidity stress testing framework.

Presenting Optimised Portfolios to Investors

Managers frequently show prospective investors an optimised allocation, a model portfolio or a backtest of a strategy blended with a typical institutional book. Under rule 206(4)-1 under the US Investment Advisers Act, the SEC marketing rule, hypothetical performance means performance not actually achieved by any portfolio of the adviser. The definition expressly includes performance derived from model portfolios and performance backtested by applying a strategy to prior periods when it was not used.

A registered adviser may include hypothetical performance in an advertisement only if three conditions are met. It must adopt and implement policies and procedures reasonably designed to ensure the performance is relevant to the likely financial situation and objectives of the intended audience. It must provide enough information for the audience to understand the criteria and assumptions used. It must also provide enough information on the risks and limitations of relying on it, which for private fund investors may be provided or offered promptly. Its application to offshore managers is set out in our SEC marketing rule guide for offshore fund managers.

The disclosures the rule requires are, in substance, the weaknesses this article has described: which period the inputs were estimated over, whether returns are arithmetic or geometric, gross or net, and how sensitive the weights are to the estimates. A frontier chart that omits them is weak analysis and, for a registered adviser, a compliance exposure.

Where the optimiser runs inside the strategy, as in many quantitative and multi-asset funds, the same discipline belongs in the fund documents and governance. CIMA's Rule on the Contents of Offering Documents for regulated mutual funds requires the offering document to include the investment policy, an indication of the techniques and instruments that may be used and a description of the fund's material risks. For an optimisation-driven strategy, estimation error and regime change are likely to be material risks. The input estimation method, the constraints and the rebalancing rules are part of the technique. Model governance for systematic strategies is covered in our guide to setting up a quantitative hedge fund.

Common Mistakes

• Treating a full-sample average as the expected return, without showing how the frontier changes across sub-periods.
• Combining gold before and after 1971, or any series that spans a change in market structure, as if it were one asset.
• Using arithmetic means to illustrate long-run wealth, when the geometric mean governs compounding: 10.02 per cent rather than 11.85 per cent for US equities over 1928 to 2025.
• Placing a fund with smoothed monthly returns on a frontier without testing for serial correlation.
• Presenting an optimised model portfolio to prospective investors without the criteria, assumptions and limitations the marketing rule requires.
• Offering liquidity terms that do not allow the rebalancing the allocator's model assumes.

The CV5 Implementation Perspective

CV5 Capital is the Cayman-headquartered institutional fund infrastructure platform for hedge fund and digital asset managers who need to launch quickly, operate properly and satisfy serious investors from day one. It operates CV5 SPC for traditional strategies and CV5 Digital SPC for digital asset strategies, both registered with CIMA under the Mutual Funds Act. Each manager receives a segregated portfolio within an SPC governed by an independent board, with coordinated fund administration, audit, banking, custody and compliance arrangements. CV5 Capital is not the investment manager of the strategies on its platforms.

For a manager whose fund will be judged on its contribution to an institutional portfolio, the Cayman hedge fund platform for emerging managers provides the administered NAV, independent valuation oversight and reporting from which allocators estimate risk and correlation. The manager retains the strategy, the portfolio construction process and every investment decision.

Key Takeaways

• Use the efficient frontier for what it shows reliably: diversification lowers risk when correlations are low, and the minimum-variance mix is dominated by the lowest-volatility asset.
• Rebuild any frontier across sub-periods before relying on it, and exclude or adjust series that span a structural break.
• Report how sensitive the optimal weights are to the inputs, using resampling or ranges rather than a single answer.
• Test and disclose serial correlation in the fund's own returns, and present risk on an unsmoothed basis where it matters.
• Where an SEC-registered adviser advertises the performance of an optimised or backtested portfolio, treat it as hypothetical performance and provide the required criteria, assumptions and limitations.
• Set dealing terms the underlying portfolio can support, because the allocator's model assumes it can rebalance.

Frequently Asked Questions

What is the efficient frontier?

The efficient frontier is the set of portfolios offering the highest expected return for each level of risk, measured as volatility. Portfolios below it are dominated, because another mix offers more return for the same risk. The concept comes from Harry Markowitz's 1952 paper Portfolio Selection in the Journal of Finance.

How is the efficient frontier calculated?

Estimate the expected return of each asset and the covariance matrix. Portfolio return is w · mu and portfolio volatility is the square root of w transpose C w. For a series of target returns, minimise variance subject to the weights summing to one, the target return being met and any constraints such as no short selling.

Which weight minimises risk for two uncorrelated assets?

The weight in the first asset equals the second asset's variance divided by the sum of the two variances. On 1928 to 2025 US data, with equity volatility of 19.40 per cent and bond volatility of 7.90 per cent, the minimum-risk mix holds about 14.2 per cent in equities.

Why do efficient frontier weights change so much over time?

Expected returns, volatilities and correlations are estimated from history and move between regimes. On US annual data the stock and bond correlation was minus 0.65 from 2000 to 2021, yet both fell about 18 per cent in 2022. Optimisers also amplify estimation error, so small input changes produce large weight changes.

Can a hedge fund be placed on an efficient frontier?

Yes, but with adjustments. Hedge fund returns are often smoothed by illiquid marks, track records are short, return distributions can be skewed, and lock-ups or gates restrict rebalancing. Allocators typically unsmooth returns, use net figures at the terms offered and add drawdown and stress measures to volatility.

Is an optimised model portfolio hypothetical performance under the SEC marketing rule?

Yes, where an SEC-registered adviser shows its performance in an advertisement. Rule 206(4)-1 defines hypothetical performance to include performance derived from model portfolios and backtested performance. A registered adviser may show it only with policies ensuring relevance to the audience and with enough information on the criteria, assumptions, risks and limitations involved.