Skip to content

New to Omni?·Get 20% off your first purchase with code WELCOME20 at checkout.See plans

CAIA® Level II: how does mean–variance choose a portfolio?

Mean–variance analysis compares portfolios through expected return and covariance. Work through an allocation, then test what the inputs and constraints hide.

ConceptOmni Curriculum Team7 min read

Two different arrangements of unmarked blue, white and wood blocks sit on a navy desk beside a blank sheet and pencil.
On this page

Mean–variance optimisation chooses among feasible portfolio weights by comparing expected return with variance, which captures how the assets move together. The answer depends on the assumed returns, covariances and constraints; it is not a mechanically correct allocation independent of those inputs. For CAIA® Level II, learn to build the comparison, recognise an efficient choice and explain why an attractive model result may need practical limits.

Define the choice before solving it

Begin with the investment universe and the rules. A portfolio weight says what fraction of capital is assigned to each asset; the weights must sum to the total invested amount. If the mandate prohibits short positions, each risky weight must be non-negative. If an illiquid sleeve has a maximum permitted weight, that bound belongs in the optimisation itself. A numerical optimum that violates the mandate is not a feasible answer, however good its expected return looks.

Expected portfolio return is a weighted average of the asset return assumptions. Portfolio variance is different: it includes each asset’s own variance and cross-terms that reflect covariance between pairs. Consequently, two assets with identical standalone volatility can contribute very differently to a portfolio. If one tends to rise and fall with the existing holdings, it offers little diversification; if its returns are less aligned, it may reduce the combined variance. The candidate should interpret covariance rather than compare volatilities in isolation.

The CAIA Association’s strategic allocation discussion describes the difficulties of applying traditional mean–variance methods to illiquid alternatives. Historical series can appear unusually smooth and correlations can be depressed by reporting lags. That can make an illiquid asset look more diversifying than it really is. The Level II curriculum resource page points candidates to the current official companion materials; this article focuses on the core calculation and interpretation rather than reproducing those materials.

What the efficient frontier means

Imagine plotting every feasible portfolio using expected return on one axis and volatility on the other. A portfolio is inefficient if another feasible mix offers at least as much expected return with less risk, or more expected return for the same risk. The efficient frontier contains the best available trade-offs under the stated inputs and rules. It is a set of conditional choices, not a single recommendation. A more risk-averse decision maker may select a different point from a less risk-averse one while both use the same opportunity set.

One formulation fixes a target return and seeks the feasible mix with minimum variance. Another assigns a positive penalty to variance and maximises expected return less that penalty. A higher penalty moves the preferred point toward lower variance, all else equal. The algebra expresses a preference trade-off; the preference itself comes from the mandate, not from the asset data. Do not infer someone’s risk tolerance merely because a solver prints a particular weight vector.

The global minimum-variance mix is the feasible combination with the smallest variance regardless of its expected return. It is an anchor for the frontier, but it is not necessarily a suitable final choice. A minimum-variance solution may have too little expected return for a stated objective. Conversely, the highest expected-return asset by itself may have a risk level outside the mandate. The exam distinction is between minimising risk, meeting a return target, and optimising a particular return-risk trade-off.

An illustrative two-asset calculation

Consider an illustrative public-equity sleeve with an expected return of 8 per cent and volatility of 20 per cent, and a bond sleeve with an expected return of 4 per cent and volatility of 10 per cent. Assume their correlation is zero. These round assumptions allow the calculation to show the role of covariance clearly. First consider a fully invested, long-only mix with half the capital in each sleeve. Its expected return is half of 8 plus half of 4, or 6 per cent.

For the same mix, the equity contribution to variance is the square of one-half times the square of 0.20, giving 0.0100. The bond contribution is the square of one-half times the square of 0.10, giving 0.0025. The cross-term is zero under the stated zero-correlation assumption. Total variance is therefore 0.0125; taking its square root gives volatility of approximately 11.18 per cent. The calculation is a model result for this invented combination, not a prediction of realised risk.

Now compare a fully invested mix with one-quarter in equities and three-quarters in bonds. Its expected return is one-quarter of 8 plus three-quarters of 4, or 5 per cent. Its variance is the square of one-quarter times 0.20 squared, plus the square of three-quarters times 0.10 squared. That is 0.0025 plus 0.005625, or 0.008125, and its volatility is approximately 9.01 per cent. The second mix gives up one percentage point of assumed return while lowering modelled volatility by about 2.17 percentage points.

Neither mix is automatically the correct final allocation. If the mandate’s maximum volatility is 10 per cent under these assumptions, the half-equity mix breaches it while the quarter-equity mix is within it. If instead the mandate requires an expected return of at least 6 per cent and accepts the corresponding risk, the quarter-equity mix misses the target. State the mandate before calling either mix optimal. More importantly, a constraint can eliminate a proposed choice even if its risk-return point looks favourable on an unconstrained plot.

A zero correlation assumption did not make the portfolio riskless: both sleeves still have their own variance. If the correlation changes, the cross-term changes and so does the answer. That is why a candidate should write down the covariance assumption, check that the same time horizon underlies all inputs, and recompute rather than carrying over the previous volatility. It is also why seemingly modest measurement errors can produce a different preferred weight.

Why alternatives need a second check

An illiquid asset may report values less frequently than a public equity fund. Appraisal-based values can move gradually even when the economic value would respond more sharply to new information. If a solver consumes that smooth return series without adjustment or scepticism, it can assign the asset low volatility and low covariance with public markets. CAIA Association notes that such lagging and smoothing may overstate diversification and encourage unrealistic allocations. The lesson is to examine how the input series was constructed before trusting the point on the frontier.

Liquidity itself is not fully represented by return variance. A private investment may have capital calls, lock-ups or slow exits, while the portfolio has obligations that fall due sooner. A return-risk plot does not show whether the allocator can raise cash when needed. A practical mandate can place a cap on illiquid assets or incorporate a liquidity penalty, as CAIA Association discusses. These constraints reduce the set of feasible mixes and may move the frontier away from an unconstrained mathematical optimum.

Sensitivity analysis is an especially useful answer when expected returns are close. Ask whether a small change in a return estimate causes a large weight shift. If it does, report that instability rather than presenting the precise weights as a robust truth. Recalculate with a lower expected return for the illiquid sleeve, a higher covariance in stress, and the mandate’s actual position limits. This is decision support: it identifies which assumptions drive the choice and where more evidence would matter.

Remember that a covariance matrix is estimated from a particular sample and horizon. A quiet period may understate co-movement during a liquidity shock. Combining monthly private-market observations with daily public-market observations without aligning them can create false precision. A single historical correlation is a description of that sample, not a permanent law. The optimiser is useful because it forces assumptions into the open; it cannot repair poor assumptions by itself.

An exam-ready method

First, identify the decision variable and the feasible set: weights, full investment and any long-only or position bounds. Second, compute expected return as a weighted average. Third, compute variance from individual variances and covariance terms, then take the square root only if the question asks for volatility. Fourth, compare feasible mixes against the stated target or risk preference. Finally, explain one input limitation that could alter the result. This sequence prevents a polished but irrelevant calculation.

Common errors are to use a simple weighted average of asset volatilities as portfolio volatility, to omit the covariance term, or to treat a lower standalone volatility as a guaranteed diversification benefit. Another is to confuse the global minimum-variance portfolio with the return-maximising portfolio. A final error is to accept a large model weight in an illiquid sleeve without checking the return data and liquidity constraint. The CAIA programme page and Omni methodology offer related context; the crucial exam habit is to articulate why the computed choice is conditional.

Sources

  1. Strategic Asset Allocation: Practical Considerations for Alternative Investments | CAIA Association · caia.org · Retrieved
  2. Curriculum and Study Tools | CAIA Association · caia.org · Retrieved

Explore CAIA study resources

Educational content, not investment advice. This article is written to help candidates prepare for professional exams. It is not a recommendation to buy, sell or hold any security, and it does not take your personal circumstances into account.

Omni Finance Academy is an independent exam-preparation provider and is not affiliated with or endorsed by CFA Institute, GARP, CAIA Association or CFP Board. CFA®, FRM®, CAIA® and CFP® are trademarks of their respective owners. Full disclaimer.