Portfolio and risk · Portfolio theory
Efficient frontier: the best portfolios and why they are hard to find
What the efficient frontier is, how it is built from eight funds over 2005 to 2026, and what happened when the best mix of one decade was held through the next.
On this page
The efficient frontier is the set of portfolios that offer the highest expected return for each level of risk. Plot every possible mix of a group of assets by volatility and return, and the mixes fill a region. Its upper-left edge is the frontier. A portfolio on that edge is called efficient: no other mix of the same assets has more return for the same risk, or less risk for the same return.
The frontier is the practical output of Modern portfolio theory: why the mix matters more than the parts, and it is also where the theory meets its main difficulty. Drawing the frontier for the past is easy. The frontier that matters is next decade's, and it has to be built from estimates. This lesson draws one from eight funds, identifies its two landmark portfolios, and then tests what an investor who trusted it would have earned.
The idea
Take a set of assets. Any choice of weights gives a portfolio with some volatility and some return. Try enough weights and a shape appears.
Each small dot is one random mix of the eight funds. No dot lies above or to the left of the curve: that region is unreachable with these assets. Dots well inside the cloud are inefficient, in the sense that another mix had more return for the same risk.
Most single funds sit inside the cloud, to the right of the frontier. Each carries volatility that a mix would have diversified away. The exception is the fund with the highest return (QQQ here), which is always the frontier's top end, because the only way to get the highest possible return is to hold nothing else.
How it works
Building a frontier takes three inputs for the assets: an expected return for each, a volatility for each, and a correlation for each pair. With eight funds that is 8 returns, 8 volatilities and 28 correlations. An optimizer then searches the weights.
The minimum-variance portfolio is the frontier's left tip. It needs no return forecasts, only volatilities and correlations, which makes it the more reliable of the two to estimate.
The highest-Sharpe portfolio is also called the tangency portfolio, because a straight line drawn from the risk-free rate just touches the frontier there. James Tobin's 1958 result is that an investor who can hold cash, or borrow, does best by holding that one portfolio and adjusting risk by mixing it with cash: all investors hold the same risky mix and differ only in how much of it. The line is called the capital market line, and it is the starting point of the Capital asset pricing model (CAPM): what return a beta should earn.
A worked example on eight funds
The data: 262 monthly total returns of each fund, January 2005 to October 2026, measured above the Treasury bill rate.
| Fund | Average return above bills | Volatility | Sharpe ratio |
|---|---|---|---|
| QQQ (Nasdaq-100) | 13.8% | 18.4% | 0.75 |
| SPY (US stocks) | 9.6% | 14.8% | 0.65 |
| GLD (gold) | 9.3% | 17.3% | 0.54 |
| IWM (small companies) | 8.2% | 19.8% | 0.42 |
| VNQ (real estate) | 7.1% | 21.5% | 0.33 |
| EFA (foreign stocks) | 5.5% | 16.6% | 0.33 |
| AGG (US bonds) | 1.2% | 4.5% | 0.27 |
| TLT (long Treasuries) | 2.1% | 13.6% | 0.15 |
Searching the weights gives:
| Portfolio | Weights | Return above bills | Volatility | Sharpe ratio |
|---|---|---|---|---|
| Minimum variance | 98% AGG, 2% SPY | 1.4% | 4.5% | 0.31 |
| Highest Sharpe | About 54% QQQ, 38% GLD, 9% TLT | 11.1% | 12.2% | 0.90 |
| Equal weight | 12.5% in each of the eight | 7.1% | 11.2% | 0.64 |
The highest-Sharpe mix had a ratio of 0.90, well above the best single fund's 0.75. It earned 11.1% above bills with 12.2% volatility, more return than SPY with less risk.
Look at what it holds. Three funds out of eight, with over half in one of them. SPY, the most widely held fund in the group, gets a weight of zero, because QQQ did the same job with a better result in this sample. The optimizer has no notion of prudence. It finds whatever combination scored best and commits to it fully.
Using it: does the best mix stay the best?
A frontier drawn from 2005 to 2026 could only be used from 2026 onwards. To see how such advice performs, split the history. Find the highest-Sharpe mix using only January 2005 to December 2015, then hold it from January 2016 to October 2026.
The two answers have little in common. The first decade's best mix was 70% bonds. The second decade's had none.
| Held from January 2016 to October 2026 | Return a year (CAGR) | Volatility | Sharpe ratio | Deepest fall |
|---|---|---|---|---|
| The 2005–2015 optimum (22% QQQ, 55% AGG, 15% TLT, 8% GLD) | 6.2% | 7.7% | 0.52 | −21.1% |
| Equal weight in the eight funds | 9.4% | 11.2% | 0.66 | −23.9% |
| 60% SPY / 40% AGG | 9.6% | 9.9% | 0.74 | −20.0% |
| The 2016–2026 optimum, known only afterwards | 16.8% | 13.1% | 1.08 | −22.3% |
In its own decade the first optimum had a Sharpe ratio of 0.98. Carried forward, it managed 0.52, behind both an equal-weight portfolio that required no analysis and a conventional 60/40 mix.
The reason is in the inputs. Over 2005 to 2015, long Treasuries earned 6.2% a year above bills and the aggregate bond fund 2.8%, as interest rates fell. Over 2016 to 2026 they earned −2.1% and −0.5%. The optimizer had loaded up on what had just done well for a reason that could not repeat. This is the pattern Richard Michaud described in 1989: an optimizer maximizes estimation error, giving the largest weights to the assets whose returns were most overestimated.
The comparison with equal weights is a well-known result. DeMiguel, Garlappi and Uppal tested fourteen optimization methods against equal weighting on several data sets and found none that reliably did better out of sample.
What practitioners do about it
- Constrain the weights. Caps on any one asset and floors on others keep the optimizer from extreme answers.
- Use modest, similar return forecasts. If every asset is assumed to earn a return in proportion to its risk, the optimizer has less to chase, and the result is driven by volatilities and correlations, which are more stable.
- Prefer the minimum-variance end. It needs no return forecast.
- Start from a sensible default. Begin with market weights or equal weights and tilt slightly, instead of starting from the optimizer's output.
- Skip return forecasts altogether. Risk parity: weighting a portfolio by risk instead of by money is one such approach.
Where it fails or misleads
- A historical frontier is hindsight. It shows what would have been best. The assets on it are there because they did well, which says little about the next period.
- Small input changes move the answer a long way. Two assets with similar returns and a high correlation are nearly interchangeable to an optimizer, and a tiny change in either's expected return shifts the whole weight from one to the other.
- The frontier is flat near the top. Many quite different mixes have almost the same Sharpe ratio. The "optimal" one is distinguished from its neighbours by amounts smaller than the estimation error.
- Volatility is the only risk on the chart. Drawdowns, illiquidity and the shape of the return distribution do not appear.
- Constraints change it. The frontier here allows no borrowing and no short sales. With either, it moves, and the highest-Sharpe portfolio can be scaled up along the capital market line.
- It ignores taxes and trading costs, which can exceed the small gains from fine-tuning.
Compared with its neighbours
| Approach | Needs return forecasts? | Needs correlations? | Typical result |
|---|---|---|---|
| Highest-Sharpe portfolio | Yes | Yes | Concentrated; unstable from period to period |
| Minimum variance | No | Yes | Dominated by the least volatile assets |
| Risk parity | No | Partly | Large weights in bonds; usually uses leverage |
| Equal weight | No | No | Spread evenly; hard to beat out of sample |
| Market weight | No | No | What all investors hold in aggregate |
How Finabase does it
Finabase does not draw an efficient frontier and has no optimizer. It does not estimate expected returns, and it does not compute correlations between holdings.
A portfolio page shows what the weights currently are: each position's market value and an allocation bar of the largest holdings and cash. The comparison chart, with several funds overlaid as percent changes from a common start, shows the raw material of a frontier: which holdings rose together and which did not over the chosen range. The comparison table lists each symbol's price return over periods up to two years, without dividends.
Anyone who wants to experiment can export month-end prices for a handful of funds and use a spreadsheet's solver on the formulas in Modern portfolio theory: why the mix matters more than the parts. The exercise is worth doing once, mainly to see how much the answer changes when the date range does.
Frequently asked questions
What does it mean for a portfolio to be "efficient"?
That no other combination of the same assets has a higher expected return for the same volatility. It is a statement about a set of assets and a set of estimates, not an absolute quality.
Is my portfolio on the efficient frontier?
Measured over the past, almost certainly not, and neither is anyone else's: the historical frontier is known only afterwards. A broadly diversified portfolio is usually not far inside it, and the distance is a poor guide to the future.
What is the tangency portfolio?
The portfolio on the frontier with the highest Sharpe ratio. A line from the risk-free rate touches the frontier at that point. In theory every investor should hold it, mixed with more or less cash.
Why did equal weighting beat the optimized portfolio?
Because the optimized weights were fitted to one decade's returns, which did not persist. Equal weighting makes no forecast, so it has no forecast error. It is not always better, but it is hard to beat reliably.
Can the frontier include borrowing or short selling?
Yes. Allowing negative weights or leverage enlarges the set of possible portfolios and moves the frontier up and to the left. It also makes the optimizer's extreme answers more extreme.
Summary
The efficient frontier is the upper-left edge of all the portfolios that can be built from a set of assets: the most return for each level of risk. For eight funds over 2005 to 2026, its highest-Sharpe point was a concentrated mix of QQQ, gold and long Treasuries with a ratio of 0.90. The frontier is exact for the past and an estimate for the future, and the estimate is fragile: the best mix of 2005 to 2015, held through 2016 to 2026, did worse than equal weights. The idea is a sound picture of the trade-off between risk and return. Its optimized output needs constraints and scepticism before it is used.
Check your understanding
4 questions on this lesson. Each answer comes with the reason, and nothing is scored.
On a chart of volatility against expected return, where does the efficient frontier lie?
- Along the lower-right edge of the possible portfolios
- Along the upper-left edge: the most return for each level of risk
- Through the middle of the cloud
Show the answer
B. Along the upper-left edge: the most return for each level of risk Efficient portfolios have the highest return available at their level of volatility, so they form the upper-left boundary of everything that can be built from the assets.
The highest-Sharpe mix of eight funds for 2005 to 2026 held only three of them and gave SPY a weight of zero. What does that illustrate?
- SPY is a poor investment
- Optimizers concentrate in whatever scored best in the sample, with no regard for prudence
- The calculation contained an error
Show the answer
B. Optimizers concentrate in whatever scored best in the sample, with no regard for prudence An optimizer maximizes one number on past data. Assets that were slightly worse than a close substitute get nothing, which produces concentrated portfolios that describe the sample.
The mix with the highest Sharpe ratio over 2005 to 2015 was 70% bonds. Held over 2016 to 2026, it trailed an equal-weight portfolio. What is the main reason?
- Bond returns in the first decade came from falling interest rates and did not persist
- Equal weighting always wins
- The Sharpe ratio stopped working
Show the answer
A. Bond returns in the first decade came from falling interest rates and did not persist The optimizer gave its largest weights to the assets with the highest past returns. Those returns were period-specific, so the weights were fitted to conditions that changed.
Which landmark portfolio can be estimated without any forecast of returns?
- The highest-Sharpe (tangency) portfolio
- The minimum-variance portfolio
- Neither
Show the answer
B. The minimum-variance portfolio Minimum variance depends only on volatilities and correlations, which are more stable than average returns. The tangency portfolio needs expected returns as well.
Sources and further reading
- Harry Markowitz, “Portfolio Selection,” Journal of Finance, vol. 7, no. 1 (1952)
- James Tobin, “Liquidity Preference as Behavior Towards Risk,” Review of Economic Studies, vol. 25, no. 2 (1958)
- Richard O. Michaud, “The Markowitz Optimization Enigma: Is ‘Optimized’ Optimal?” Financial Analysts Journal, vol. 45, no. 1 (1989)
- Victor DeMiguel, Lorenzo Garlappi and Raman Uppal, “Optimal Versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy?” Review of Financial Studies, vol. 22, no. 5 (2009)
- Finabase chart data — monthly bars and dividends of SPY, QQQ, EFA, IWM, AGG, TLT and VNQ, and monthly bars of GLD, January 2005 to October 2026; statistics computed by Finabase
