Modern Portfolio Theory and the Efficient Frontier
5 steps · one page
In short
Once risk depends on correlation, a question follows immediately: of all the ways to combine a set of holdings, which combinations are not wasteful?
Scope. This article explains what the efficient frontier is, how the tangency portfolio is derived, and why the machinery is far less usable than it appears. It recommends no allocation and suggests no weights. The portfolios computed below exist to demonstrate the method and its instability — none of them is offered as a portfolio anyone should hold. All figures use the illustrative teaching parameters fixed on this pillar's hub and are not forecasts of any market.
A combination is wasteful if another exists with the same variability and a higher expected return, or the same expected return and lower variability. Remove all the wasteful ones and what remains is the efficient frontier — the set of combinations where no improvement is available without accepting something worse elsewhere.
That is the whole of the concept, and it is genuinely one of the most important results in finance. The difficulty is not the idea. It is what happens when anyone tries to use it.
Building it
On the hub parameters — equities at a 9.0% expected return with 16.0% variability, bonds at 5.0% and 6.0%, correlation 0.20, and a risk-free rate of 4.0%.
| Portfolio | Equities | Expected return | Standard deviation | Return per unit of risk above the risk-free rate |
|---|---|---|---|---|
| All bonds | 0.0% | 5.00% | 6.00% | 0.1667 |
| Minimum variance | 6.6% | 5.26% | 5.91% | 0.2140 |
| Tangency | 50.1% | 7.00% | 9.10% | 0.3301 |
| All equities | 100.0% | 9.00% | 16.00% | 0.3125 |
Worked example
Worked example — two special points, and what each one is. The minimum-variance portfolio is the least variable combination available: 6.6% in equities at a standard deviation of 5.91%, which as the first article showed is below holding bonds alone. The tangency portfolio is the combination with the highest return per unit of risk above the risk-free rate: 50.1% in equities, scoring 0.3301. Note that it beats holding equities outright — 0.3301 against 0.3125 — despite having a much lower expected return, because it gives up 2.00 points of return to remove 6.90 points of variability. The theory's central conclusion follows: once a risk-free asset exists, every investor's risky holdings should in principle be the same combination, with only the split between that combination and the risk-free asset varying by preference. That conclusion is elegant, it is what the mathematics says, and the next section is about why it does not survive contact with real inputs.
Why the machinery does not work as advertised
The frontier is computed from expected returns, variabilities and correlations. All three are estimates, and the optimiser treats them as facts.
The consequence is more severe than ordinary estimation error, and it has a name: error maximisation. An optimiser searching for the best combination will systematically favour whichever assets have the most overstated expected returns and the most understated correlations — because that is exactly what "looks best" means to it. The procedure does not merely tolerate estimation error; it seeks it out.
The scale of the problem can be shown by varying one input by a small amount and watching the answer move.
| Assumed expected return on equities | Tangency portfolio's equity weight |
|---|---|
| 8.0% | 41.1% |
| 8.5% | 45.7% |
| 9.0% (hub parameter) | 50.1% |
| 9.5% | 54.3% |
| 10.0% | 58.3% |
Worked example — a one-point error in a single input moves the answer by nine points. Shifting the assumed equity return from 9.0% to 10.0% moves the optimal equity weight from 50.1% to 58.3%; shifting it down to 8.0% moves it to 41.1%. A two-point range in one assumption produces a seventeen-point range in the recommended holding. And a two-point range on an equity return estimate is not pessimistic — it is narrower than the disagreement among serious practitioners, as Pillar 27's article on the cost of capital shows for a closely related estimate. Three further points make it worse rather than better: this is a two-asset problem, and instability grows sharply with the number of assets; correlations are estimated with more error than returns and are unstable in the direction that hurts; and the optimiser gives no indication that its answer is fragile — it returns 50.1% with the same confidence whatever the inputs deserve. This portal publishes no optimiser and no frontier tool, and this table is the reason.
What survives
The criticism above is aimed at the optimiser, not at the theory, and the distinction matters. Three things from Markowitz's framework are robust and are used throughout this pillar.
That risk is a property of the collection rather than of holdings. This is a mathematical fact and does not depend on estimating anything precisely.
That combinations exist which dominate others. The existence of wasteful portfolios is not in doubt even if their precise identification is.
That the relevant question about any holding is what it does to the whole. This survives every objection and reframes the entire subject.
Worked example
The honest summary, and it is the pillar's position. Modern portfolio theory is a superb conceptual framework and a poor calculating device. As a way of thinking it establishes that diversification has a mathematical basis, that holdings must be judged in context, and that there is such a thing as an unnecessarily inefficient portfolio. As a machine for producing weights it converts small errors in unknowable inputs into large differences in output, without signalling that it has done so. The framework's own author identified the input problem, and it has been studied continuously since without being solved — because it is not a flaw in the mathematics but a consequence of the mathematics being exact about quantities that are not.
Frequently asked
8 questions
What is the efficient frontier?
The set of combinations for which no alternative offers the same variability with a higher expected return, or the same expected return with lower variability. Everything else is wasteful by construction.
What is the minimum-variance portfolio?
The least variable combination available. On the hub parameters it holds 6.6% in equities with a standard deviation of 5.91% — below holding bonds alone.
What is the tangency portfolio?
The combination with the highest return per unit of risk above the risk-free rate — 50.1% in equities on these parameters, scoring 0.3301 against 0.3125 for equities outright, because it gives up 2.00 points of return to remove 6.90 points of variability.
Why does the theory say everyone should hold the same risky combination?
Because once a risk-free asset exists, the highest-scoring risky combination is the same for everyone, and only the split between it and the risk-free asset varies with preference. That is what the mathematics says on given inputs; whether the inputs can be known is a different question.
What is error maximisation?
An optimiser systematically favours the assets whose expected returns are most overstated and whose correlations are most understated, because that is what looks best to it. It does not merely tolerate estimation error — it seeks it out.
How unstable is the output?
Severely. Moving the assumed equity return from 9.0% to 10.0% shifts the optimal equity weight from 50.1% to 58.3%; moving it to 8.0% shifts it to 41.1%. A two-point range in one assumption produces a seventeen-point range in the answer, in a two-asset problem where instability is at its mildest.
Does that mean the theory is wrong?
No — the criticism is aimed at the optimiser. That risk is a property of the collection, that some portfolios dominate others, and that a holding must be judged by what it does to the whole are all robust and do not depend on estimating anything precisely.
Why does MarketClue not offer an optimiser?
Because it would present the most fragile output in finance with the appearance of precision. The optimiser returns a weight with the same confidence whatever the inputs deserve, and gives no indication that its answer is fragile.
References
- Markowitz (1952) — Portfolio Selection, Journal of Finance 7(1) —
- Michaud (1989) — The Markowitz Optimization Enigma: Is "Optimized" Optimal?, Financial Analysts Journal 45(1) (error maximisation) —
- Tobin (1958) — Liquidity Preference as Behavior Towards Risk, Review of Economic Studies 25(2) (the separation result) —
Educational and informational only — not investment advice, a recommendation, or an offer to buy or sell any security. Investing involves risk, including the possible loss of principal. Worked examples use fictional companies and figures.