Portfolio construction is the process of choosing investments and deciding how much capital to allocate to each. Its mathematics connects three questions: what return might the portfolio earn, how much might its value fluctuate, and how do the holdings behave together?
The central distinction is simple: portfolio return is a weighted average, but portfolio risk usually is not. Combining investments can reduce volatility without reducing expected return by the same proportion. That benefit depends on their relationships, not just the number of holdings.
The calculations below show how weights, expected returns, volatility and correlation fit together. All numerical assumptions are hypothetical, not market forecasts or recommended allocations.
Define the investment problem before choosing weights
An allocation needs an objective. “Maximize returns” leaves too much unanswered: over what period, with what acceptable losses, and with how much access to the money?
A portfolio intended to fund spending in two years faces a different problem from one accumulating retirement savings over three decades. Time horizon and willingness to accept losses belong in the allocation decision, alongside the return calculations. These considerations are central to the SEC’s guidance on asset allocation and diversification.
Translate the objective into boundaries before running an optimization. An illustrative mandate might prohibit borrowing, require a cash reserve and cap exposure to any individual issuer. These are inputs to the problem, not adjustments to make after an attractive result appears.
Also distinguish risk tolerance from financial capacity. An investor might feel comfortable with volatile holdings but still need the money for a scheduled payment. A spreadsheet cannot postpone that payment because markets have become inconvenient.
Once the objective and boundaries are clear, the mathematics can compare portfolios that are actually usable.
Calculate portfolio weights and expected return
A holding’s weight is its market value divided by the portfolio’s total value:
wi = Vi / Vp
Here, Vi is the holding’s value and Vp is total portfolio value. A $30,000 holding in a $100,000 portfolio has a weight of 0.30, or 30%.
For a fully invested portfolio without short positions, weights are nonnegative and sum to one. Include cash as a holding rather than leaving it outside the calculation.
Expected portfolio return is the sum of each weight multiplied by that investment’s expected return:
E(Rp) = ∑ wi × E(Ri)
This weighted relationship is part of the standard formulation in the CVX Group’s portfolio optimization model. It describes an expected return over a defined period, not a guaranteed outcome.
Suppose a portfolio holds 60% equities with an assumed annual expected return of 8%, and 40% bonds with an assumed return of 4%:
E(Rp) = (0.60 × 8%) + (0.40 × 4%) = 6.4%
The arithmetic is straightforward. Choosing defensible return assumptions is harder. An optimizer treats an optimistic forecast as an input, not as a claim that needs questioning.
This calculation also does not establish a compound annual growth rate. Compounding depends on the sequence of realized portfolio returns and how weights change between periods.
Measure risk through covariance, not weighted volatility
Volatility measures the dispersion of returns around their average. Portfolio calculations commonly express it as the standard deviation of returns, written as σ. Variance is volatility squared.
The portfolio’s variance depends on both individual asset variances and their pairwise covariances. This relationship is central to the Royal Swedish Academy’s account of Markowitz’s portfolio theory: an investment’s contribution to the combined portfolio matters more than its risk viewed alone.
For two assets, portfolio variance is:
σp2 = w12σ12 + w22σ22 + 2w1w2σ1σ2ρ12
The last term captures how the assets move together. The symbol ρ represents correlation, which ranges from −1 to +1. Covariance equals correlation multiplied by the two assets’ volatilities.
With positive weights, lower correlation reduces portfolio variance when the other inputs remain unchanged. Negative correlation is not required for diversification to help. Two assets with positive correlation below one can still produce less volatility than their weighted average volatility.
For many holdings, the compact version is σp2 = wTΣw, where w contains the weights and Σ contains all asset variances and covariances. The notation looks more intimidating than the task: account for every holding and every pairwise relationship.
The distinction between these inputs is covered further in correlation and covariance in investment portfolios. Correlation describes the relationship on a standardized scale; covariance also reflects the size of the return fluctuations.
A worked portfolio construction example
Consider a $100,000 portfolio with the following assumptions. Returns and volatilities use the same annual horizon and reporting currency.
| Input | Equities | Bonds |
|---|---|---|
| Starting value | $60,000 | $40,000 |
| Portfolio weight | 60% | 40% |
| Expected annual return | 8% | 4% |
| Annual volatility | 18% | 6% |
| Assumed equity–bond correlation | 0.20 | |
The expected return is 6.4%, as calculated above. Entering volatility and weights as decimals gives:
σp2 = (0.602 × 0.182) + (0.402 × 0.062) + (2 × 0.60 × 0.40 × 0.18 × 0.06 × 0.20) = 0.0132768
Taking the square root produces portfolio volatility of approximately 11.52%.
Simply averaging the volatilities would give 13.2%: 60% of 18%, plus 40% of 6%. That overstates volatility under the assumed correlation because it ignores the diversification benefit.
The result is conditional on the inputs. It does not mean losses stop at 11.52%, or that returns stay within 11.52 percentage points of their expectation. Standard deviation is a measure of dispersion, not a loss ceiling.
Use the efficient frontier to compare allocations
Changing the weights creates different combinations of expected return and volatility. Some combinations are inferior to others within the model.
If portfolio A has the same expected return as portfolio B but lower volatility, B offers no advantage under those two criteria. The efficient frontier consists of portfolios offering the highest expected return for a given volatility, or the lowest volatility for a given expected return.
The efficient frontier and its allocation trade-offs provide a framework for making that comparison. There is no single best point for every investor: the choice depends on the objective and constraints.
The frontier also belongs to a particular model. Change expected returns, the covariance matrix, available investments or borrowing restrictions, and the boundary can move.
A minimum variance portfolio answers a narrower question: which permitted allocation has the lowest estimated variance? It does not automatically provide enough expected growth to meet a spending goal. Low modeled volatility and financial suitability are separate tests.
Separate capital allocation from risk allocation
A portfolio that splits money evenly does not necessarily split risk evenly. Holdings with higher volatility or stronger relationships with the rest of the portfolio can account for much more of its fluctuations.
In the worked example, equities represent 60% of capital but approximately 92% of modeled portfolio volatility under the standard covariance based contribution calculation. Bonds represent 40% of capital yet contribute only about 8% of that volatility.
This distinction motivates risk parity and allocation by risk contribution. Rather than choosing equal monetary weights, a risk budgeting approach sets targets for how much risk each holding contributes.
That is a different objective, not proof of a better portfolio. Equal risk contributions do not imply equal expected returns, protection from losses or an appropriate allocation for every investor. The method still depends on estimated relationships between investments.
Test the assumptions before trusting the output
Prepare comparable return data
Use observations covering the same dates, frequency and reporting currency. A covariance calculation built from mismatched periods answers the wrong question, even if the software accepts it.
Distinguish price returns from total returns that include investment income. Also identify whether the series already reflects product expenses. Otherwise, income may be omitted or costs deducted twice.
Keep the units consistent. Monthly volatility and annual volatility are not interchangeable, and 18% must enter a decimal based formula as 0.18 rather than 18.
Examine estimation error and adverse scenarios
Future expected returns and covariances are unknown. Historical estimates contain sampling error, and an allocation optimized around those estimates can perform poorly when the estimates are wrong. This problem is examined directly in research on mean–variance optimization with unknown means and covariances.
Test the allocation under alternative assumptions rather than treating one forecast as settled. Lower expected returns, increase volatility, or change correlations. Watch whether modest input changes produce large changes in recommended weights.
In the worked example, increasing correlation from 0.20 to 0.80 while leaving everything else unchanged raises modeled volatility from 11.52% to approximately 12.80%. That is a scenario calculation, not a prediction.
Testing what happens when correlations rise during market stress helps reveal how much diversification the allocation depends on. A stress scenario should also consider simultaneous losses, rather than changing correlation alone.
Include costs and changes in portfolio weights
A portfolio should be evaluated after its expected implementation costs. Product expenses, trading costs and any borrowing charges can reduce the return available to the investor.
These deductions also reduce the capital left to earn future returns. The cumulative effect is illustrated in the SEC bulletin on how fees affect investment portfolios. An apparently better allocation may offer little benefit if maintaining it requires expensive trading.
For illustration, an annual portfolio expense estimated at 0.4% would reduce the example’s 6.4% gross expected return to roughly 6.0%, before other costs and taxes. This subtraction is an approximation; actual deductions depend on their timing and calculation method.
Weights also change after construction. If the $60,000 equity allocation gains 10% while the $40,000 bond allocation is unchanged, total value becomes $106,000. Equities then represent approximately 62.26% of the portfolio, not 60%.
That drift changes the next period’s exposure. The investor must decide whether to restore the original weights, accept the change or revise the target because the objective has changed.
A written portfolio rebalancing rule using dates or allocation thresholds makes that decision repeatable. Contributions can also be directed to underweight holdings, although restoring a target is not always possible through contributions alone.
Judge the portfolio, not the decimal places
A useful construction process starts with the investor’s objective, calculates comparable return and risk estimates, applies practical constraints, then tests whether the allocation remains acceptable under less favorable assumptions.
Keep a simple benchmark for comparison. A more complicated allocation needs to justify its added costs, trading and dependence on forecasts.
Portfolio mathematics makes assumptions visible and trade-offs measurable. It does not remove uncertainty. The useful output is not the most precise set of weights, but an allocation whose behavior, weaknesses and maintenance requirements are clear enough to act on.