Modern Portfolio Theory and Diversification

Introduction

Investment decisions cannot be evaluated solely by identifying securities with the highest expected returns. Returns are uncertain, and the risks of securities interact when they are combined. Modern Portfolio Theory (MPT), developed principally by Harry Markowitz, provides a systematic framework for selecting portfolios by considering expected return, variance and covariance simultaneously.

The central insight is that portfolio risk depends not only on each asset’s individual volatility but also on how asset returns move together. Consequently, a risky security may improve a portfolio when its return has sufficiently low covariance with the returns of existing holdings. Diversification is therefore a portfolio-level decision rather than merely the ownership of many securities.

This chapter develops mean–variance analysis, efficient portfolios, the global minimum-variance portfolio, investor indifference curves and the optimal risky portfolio. It also distinguishes systematic from unsystematic risk and critically examines short-selling assumptions, estimation risk and practical limitations of portfolio optimisation.

Learning Objectives

After studying this chapter, students should be able to:

1. Calculate portfolio expected return, variance and standard deviation.
2. Explain covariance, correlation and the economic basis of diversification.
3. Derive and interpret two-asset and multi-asset minimum-variance portfolios.
4. Distinguish feasible, minimum-variance and efficient portfolios.
5. Explain investor risk aversion using mean–variance indifference curves.
6. identify the tangency portfolio and determine the allocation between risky and risk-free assets.
7. Distinguish systematic risk from diversifiable unsystematic risk.
8. Evaluate short-selling assumptions and practical optimisation constraints.
9. Apply portfolio theory in an investment-management setting.

1. The Mean–Variance Framework

1.1 Returns as uncertain outcomes

The holding-period return on an asset is:

where R_i is the asset’s return over the period, P_0,i is its initial price, P_1,i is its ending price and D_i is income received, such as dividends or interest. All monetary variables must use the same currency, while return is dimensionless and is normally expressed as a decimal or percentage.

Because the future selling price and income may be uncertain, return is treated as a random variable. If possible returns R_i,s have probabilities p_s, expected return is:

where μ_i is expected return, S is the number of mutually exclusive states and ∑p_s = 1.

Variance and standard deviation measure dispersion around expected return:

Variance is measured in squared-return units, whereas standard deviation has the same units as return. Mean–variance analysis uses variance as a risk measure and is most defensible when returns are approximately elliptical, investors have quadratic utility, or mean and variance adequately represent preferences. It does not fully describe skewness, tail dependence or extreme loss.

1.2 Portfolio expected return

For n assets, portfolio return is the weighted sum of asset returns:

The portfolio weights w_i represent proportions of net portfolio wealth and normally satisfy:

If short selling is prohibited, w_i ≥ 0. When short selling is allowed, some weights may be negative, while long positions may exceed 100%.

By linearity of expectation:

Expected portfolio return is therefore a weighted average even when asset returns are correlated.

2. Covariance, Correlation and Portfolio Risk

2.1 Covariance and correlation

Covariance measures the joint movement of two returns:

A positive covariance indicates a tendency to move together; a negative covariance indicates offsetting movements. Its magnitude depends on the scale of the variables. Correlation standardises covariance:

provided σ_i > 0 and σ_j > 0. Correlation satisfies −1 ≤ ρ_ij ≤ 1, and therefore:

Correlation does not imply causation and may change substantially during market stress.

2.2 Two-asset portfolio variance

For two assets, A and B, with weights w_A and w_B = 1 − w_A, portfolio variance is:

Equivalently:

The final term explains why portfolio risk is generally not a weighted average of individual standard deviations.

Worked Example 1: Diversification with two assets

Suppose Fund A has expected return 10% and standard deviation 15%, while Fund B has expected return 6% and standard deviation 8%. Their correlation is 0.25. A portfolio invests 60% in A and 40% in B.

Expected return is:

Covariance is:

Portfolio variance is:

Thus:

The weighted average of the individual standard deviations is 12.2%, but actual portfolio standard deviation is approximately 10.28%. The difference is the diversification benefit generated by imperfect correlation.

2.3 Correlation cases

With perfect negative correlation, the zero-risk long-only weights are:

This is a special two-asset result and must not be generalised to correlations other than −1.

3. Multi-Asset Portfolio Risk

Let w be an n × 1 weight vector, μ the expected-return vector and Σ the n × n covariance matrix. Then:

The covariance matrix is symmetric. A valid population covariance matrix is positive semidefinite; optimisation commonly assumes it is positive definite so that it is invertible and every non-zero risky position has positive variance.

Expanded portfolio variance is:

This expression shows that adding assets increases the importance of covariance estimation. For n assets, there are n(n−1)÷2 distinct pairwise covariances.

4. Diversification and the Sources of Risk

4.1 Systematic and unsystematic risk

Under a single-index market model:

where R_M is the market return, β_i measures sensitivity to market movements, α_i is the intercept and ε_i is the asset-specific residual. If Cov(R_M,ε_i)=0, then:

The first component is systematic risk arising from economy-wide influences such as interest rates, recessions, inflation and geopolitical shocks. The second is unsystematic risk associated with firm- or industry-specific events.

If residual returns are mutually uncorrelated and n assets have equal weights and identical residual variance σ_ε², portfolio residual variance is:

This decline toward zero is a special result requiring equal weights and uncorrelated residuals. In practice, residual correlations, concentrated weights and common industry exposures prevent complete elimination. Systematic risk remains because securities are jointly exposed to market-wide factors.

4.2 Effective rather than nominal diversification

A portfolio containing many securities may still be poorly diversified if those securities share similar exposures. Effective diversification requires attention to:

• economic sectors and business models;
• countries and currencies;
• duration and interest-rate sensitivity;
• credit quality and default factors;
• liquidity and funding risk;
• inflation, commodity and market-beta exposures;
• nonlinear exposures created by derivatives.

Diversification may also weaken during crises because correlations often increase precisely when protection is most valuable.

5. The Feasible Set and Efficient Frontier

The feasible set contains all portfolios satisfying the applicable budget and investment constraints. A portfolio is mean–variance efficient if no other feasible portfolio:

1. has at least the same expected return and strictly lower variance; or
2. has at most the same variance and strictly higher expected return.

Thus, efficiency excludes both forms of mean–variance dominance. The efficient frontier is the upper boundary of the minimum-variance set in expected-return–standard-deviation space.

5.1 Global minimum-variance portfolio

The global minimum-variance (GMV) portfolio has the lowest variance among all fully invested risky portfolios, without imposing a target return:

When Σ is positive definite and short selling is unrestricted, the solution is:

Its variance is:

The vector 1 contains n ones. With no-short-sale or other inequality constraints, these closed-form expressions generally do not apply; quadratic programming is required.

Worked Example 2: Two-asset GMV portfolio

Using Funds A and B from Worked Example 1, the minimum-variance weight in A is:

This formula applies specifically to a fully invested two-asset portfolio. Substitution gives:

Therefore, w_B^* = 0.8515. Expected return is approximately:

The minimum variance can also be calculated as:

Hence σ_GMV ≈ 7.68%. The GMV point is the leftmost attainable portfolio in mean–standard-deviation space. The branch above it is efficient; the branch below it is inefficient.

5.2 Minimum variance for a target return

For target return μ₀, the unconstrained problem is:

Define:

If Σ is positive definite and expected returns are not all identical, then d > 0. The frontier variance is:

This equation describes the unconstrained minimum-variance frontier. Regulatory, short-sale and allocation constraints alter its shape and may produce corner solutions.

6. Risk-Free Asset, Capital Allocation Line and Tangency Portfolio

A risk-free asset has a known holding-period return r_f and zero return variance over the relevant horizon. It is assumed to have zero covariance with risky assets. In reality, instruments used as proxies can retain inflation, reinvestment, currency or sovereign risk.

Let y be the proportion invested in risky portfolio P and 1−y the proportion in the risk-free asset. Then:

For y ≥ 0, eliminating y gives the capital allocation line (CAL):

Its slope is the Sharpe ratio of portfolio P. The CAL with the greatest feasible slope is tangent to the risky-asset efficient frontier. Its point of tangency is the optimal risky or tangency portfolio.

With unrestricted short selling, a positive-definite covariance matrix and a positive normalising denominator, tangency weights are:

This portfolio maximises the ex ante Sharpe ratio. If short sales or leverage are constrained, the optimal risky portfolio must instead be found through constrained optimisation.

Figure: The GMV portfolio is the minimum-standard-deviation point. The upper branch is efficient, and the CAL is tangent to it at Q.

Worked Example 3: Tangency portfolio

For Funds A and B, let the risk-free rate be 3%. The covariance matrix and excess-return vector are:

Applying the tangency formula and normalising the resulting positions gives approximately:

The portfolio’s expected return is approximately 7.74%, variance is 0.007775 and standard deviation is 8.82%. Its Sharpe ratio is:

7. Risk Aversion, Indifference Curves and the Complete Portfolio

A commonly used mean–variance utility function is:

where γ > 0 is the investor’s coefficient of risk aversion. Returns must be entered consistently as decimals or consistently scaled percentages; changing units without adjusting γ changes the numerical utility calculation.

For a fixed utility level:

These indifference curves slope upward and become steeper as risk rises. Larger γ implies stronger risk aversion. The investor’s optimal complete portfolio occurs where the highest attainable indifference curve is tangent to the best CAL.

If borrowing and lending occur at the same risk-free rate and the investor can select any y, optimal investment in the tangency portfolio is:

If 0 < y^* < 1, some wealth is lent at the risk-free rate. If y^* = 1, all wealth is invested in the tangency portfolio. If y^* > 1, the mathematical solution requires borrowing or leverage. Institutional restrictions may force a boundary solution.

Using Worked Example 3 and γ = 4:

The unconstrained investor would place approximately 152% in the tangency portfolio and borrow about 52% at the risk-free rate. This result is not necessarily implementable because actual borrowing rates are higher than lending rates and leverage limits may apply.

8. Short Selling and Leverage

A short sale involves selling a borrowed security with the intention of repurchasing it later. In portfolio weights, a short position appears as w_i < 0. Short selling expands the feasible set and can improve the theoretical efficient frontier, but the standard model often assumes more freedom than markets provide.

Practical complications include:

• securities-borrowing fees and collateral requirements;
• margin calls and forced liquidation;
• potentially unlimited loss when a shorted security rises;
• recall risk and short-sale bans;
• dividend or interest payments owed to the lender;
• different borrowing and lending rates;
• gross-exposure and leverage limits.

A fully invested portfolio may have net exposure of 100% but much larger gross exposure. For example, weights of 140% and −40% sum to 100%, yet gross exposure equals 180%. Risk governance should therefore monitor both net and gross positions.

9. Illustrative Case: Portfolio Design for a Bank-Sponsored Fund

Background

This fictional case concerns Meridian Bank’s balanced investment fund. The investment committee is choosing between Fund A and Fund B from the preceding examples. The fund’s planning assumptions are: expected returns of 10% and 6%, standard deviations of 15% and 8%, correlation of 0.25 and a risk-free rate of 3%.

Problem

The committee initially wants an expected return of 8%, but it also requires evidence that the portfolio is diversified and consistent with client risk limits. Short selling is prohibited, and retail portfolios may not borrow.

Analysis

To achieve an 8% expected return:

Therefore, w_A = 0.50 and w_B = 0.50. Its variance is:

Thus, standard deviation is approximately 9.34%. The target portfolio has less risk than Fund A while retaining an expected return above Fund B’s return.

The unconstrained tangency portfolio contains 43.5% in A and 56.5% in B. A client with risk-aversion coefficient 4 would mathematically prefer leveraged exposure of approximately 152% to that portfolio. However, the retail mandate prohibits borrowing. The feasible allocation is therefore capped at y = 1, producing a 100% investment in the tangency portfolio rather than the unconstrained utility maximum.

Approach and outcome

The committee adopts two model portfolios:

• a target-return portfolio with 50% in each fund for mandates explicitly seeking an 8% expected return;
• the 43.5%–56.5% tangency portfolio for growth-oriented clients where maximising estimated risk-adjusted return is appropriate.

The estimates are subjected to stress tests using higher correlations and lower expected returns. Position limits and quarterly rebalancing rules are documented. The committee avoids presenting expected returns as guarantees.

Learning points

1. A target-return portfolio and the maximum-Sharpe portfolio solve different problems.
2. Mathematical leverage recommendations must be reconciled with legal and suitability constraints.
3. Diversification benefits depend on uncertain covariance estimates.
4. Governance, disclosure and stress testing are integral to professional portfolio construction.

10. Practical Limitations of Optimisation

10.1 Estimation error and instability

Expected returns, variances and covariances are unknown and must be estimated. Expected-return estimates are especially noisy. Small changes in inputs can produce large changes in optimal weights, often creating concentrated or unintuitive portfolios. This phenomenon is sometimes described as error maximisation because the optimiser may assign extreme weights to assets whose estimated returns are overstated.

Common responses include:

• shrinkage estimators for covariance matrices;
• Bayesian or Black–Litterman expected returns;
• robust optimisation over ranges of inputs;
• resampling and scenario analysis;
• weight, sector, turnover and tracking-error constraints;
• regularisation and penalties for concentrated positions.

These methods reduce instability but do not eliminate model risk.

10.2 Non-normal returns and downside risk

Variance penalises positive and negative deviations equally. Investors may instead be concerned with downside deviation, value at risk, expected shortfall, drawdown or failure to meet liabilities. Options and credit instruments may have skewed or fat-tailed return distributions, making mean–variance analysis incomplete.

10.3 Market frictions and changing conditions

Classical optimisation may omit taxes, bid–ask spreads, market impact, minimum trade sizes, illiquidity and transaction costs. Correlations and volatilities are also time-varying. Historical estimates may be misleading during structural breaks, monetary-policy changes or financial crises.

10.4 Governance, ethics and regulation

Investment professionals should ensure that optimisation outputs are suitable for the client and consistent with mandates, liquidity needs and regulatory limits. Important governance practices include independent model validation, documented assumptions, data-quality controls, stress testing, approval of overrides and transparent performance attribution.

It is unethical to portray model-generated expected returns as certain outcomes or to conceal sensitivity to assumptions. ESG or sustainability constraints should be defined transparently rather than introduced through undisclosed subjective adjustments.

10.5 Technology and contemporary practice

Modern portfolio platforms can estimate large covariance matrices, solve constrained quadratic programs and perform scenario analysis rapidly. Machine-learning methods may identify nonlinear relationships or changing regimes, but they can overfit historical data and reduce interpretability. AI-assisted optimisation should therefore remain subject to human oversight, out-of-sample testing, explainability standards and model-risk governance.

11. Professional Portfolio-Construction Process

A defensible process generally includes:

1. defining the investor’s objectives, horizon and risk capacity;
2. specifying eligible assets and legal constraints;
3. developing documented capital-market assumptions;
4. estimating and validating the covariance matrix;
5. generating feasible portfolios under realistic constraints;
6. testing sensitivity to returns, volatility and correlations;
7. incorporating costs, liquidity and taxes;
8. selecting a portfolio consistent with investor preferences;
9. implementing and rebalancing under controlled procedures;
10. monitoring exposures, outcomes and model performance.

Optimisation should support professional judgement, not replace it.

Chapter Summary

Modern Portfolio Theory evaluates investments as components of portfolios rather than in isolation. Expected portfolio return is a weighted average of expected asset returns, while portfolio variance depends on individual variances and pairwise covariances. Imperfect correlation creates diversification benefits, although systematic risk cannot be eliminated merely by increasing the number of securities.

The GMV portfolio has the smallest attainable variance. Portfolios above the GMV point on the minimum-variance boundary form the efficient frontier because they are not dominated in both expected return and risk. Introducing a risk-free asset creates capital allocation lines. The line tangent to the efficient frontier identifies the risky portfolio with the highest Sharpe ratio under the model’s assumptions.

Investor risk aversion determines the allocation between the tangency portfolio and the risk-free asset. Short-selling restrictions, leverage limits, transaction costs and unequal borrowing and lending rates can materially alter theoretical solutions. In practice, estimation error, unstable correlations, non-normal returns and model risk require constraints, stress testing, transparent governance and professional judgement.

Key Terms

Capital allocation line: The risk–return line formed by combining a risk-free asset with a particular risky portfolio.

Correlation: A standardised measure of linear co-movement between two returns.

Covariance: A measure of how two asset returns vary together.

Diversification: The combination of imperfectly correlated exposures to reduce portfolio-specific risk.

Efficient frontier: The set of feasible portfolios not dominated by another portfolio in mean–variance terms.

Expected return: The probability-weighted average of possible returns.

Feasible set: All portfolios satisfying the applicable investment constraints.

Global minimum-variance portfolio: The fully invested risky portfolio with the lowest attainable variance.

Gross exposure: The sum of the absolute values of portfolio weights.

Indifference curve: Combinations of expected return and risk providing equal investor utility.

Mean–variance analysis: Portfolio evaluation based on expected return and return variance.

Optimal risky portfolio: The risky portfolio selected for combination with the risk-free asset; under standard assumptions, it is the tangency portfolio.

Risk aversion: A preference for lower risk when expected return is held constant.

Sharpe ratio: Excess expected return per unit of total standard deviation.

Short selling: Selling a borrowed security, represented by a negative portfolio weight.

Systematic risk: Market-wide risk that cannot generally be eliminated through diversification.

Tangency portfolio: The risky portfolio at which the maximum-slope CAL touches the efficient frontier.

Unsystematic risk: Asset- or firm-specific risk that can be reduced through diversification.

Variance: Expected squared deviation of return from its mean.

Review Questions

Short Answer Questions

1. Why is portfolio standard deviation not generally a weighted average of asset standard deviations?
2. Distinguish covariance from correlation.
3. What makes a portfolio mean–variance efficient?
4. Distinguish systematic risk from unsystematic risk.
5. What does a negative portfolio weight represent?

Long Answer Questions

1. Explain the theoretical foundations and assumptions of mean–variance analysis.
2. Derive the two-asset portfolio variance equation and explain the role of correlation.
3. Explain the GMV portfolio, efficient frontier and tangency portfolio.
4. Discuss how risk aversion determines an investor’s optimal complete portfolio.
5. Critically evaluate the practical limitations of portfolio optimisation.

Analytical / Application Questions

1. Two assets have expected returns of 12% and 7%, standard deviations of 18% and 10%, and correlation of 0.30. Calculate the expected return and standard deviation of a portfolio investing 40% in the first asset.
2. Using a mean–variance utility function, analyse how the optimal risky allocation changes when risk aversion doubles, holding all market inputs constant.
3. Design a governance framework for a bank that uses AI-assisted covariance estimates in constrained portfolio optimisation.

Multiple Choice Questions with Answers

Question 1

A portfolio invests 60% in an asset expected to return 10% and 40% in an asset expected to return 5%. What is its expected return?

A. 7.0%

B. 8.0%

C. 8.5%

D. 9.0%

Correct Answer: B. 8.0%

Question 2

What is the principal portfolio effect of covariance?

A. It determines the weighted average expected return.

B. It guarantees that every additional asset reduces risk.

C. It measures how joint return movements affect portfolio variance.

D. It converts expected returns into risk-free returns.

Correct Answer: C. It measures how joint return movements affect portfolio variance.

Question 3

Which condition correctly describes a mean–variance-efficient portfolio?

A. No feasible portfolio has at least the same expected return with strictly lower variance, or at most the same variance with strictly higher expected return.

B. It has the highest expected return among all securities, regardless of risk.

C. It has lower variance than every feasible portfolio, including the GMV portfolio.

D. It contains only assets with negative pairwise correlations.

Correct Answer: A. No feasible portfolio has at least the same expected return with strictly lower variance, or at most the same variance with strictly higher expected return.

Question 4

Funds A and B have standard deviations of 15% and 8%, respectively, and correlation 0.25. What is their covariance?

A. 0.0003

B. 0.0030

C. 0.0300

D. 0.3000

Correct Answer: B. 0.0030

Question 5

What identifies the global minimum-variance portfolio in expected-return–standard-deviation space?

A. It has the greatest expected return.

B. It is the leftmost attainable risky portfolio.

C. It has the highest beta.

D. It is always invested partly in the risk-free asset.

Correct Answer: B. It is the leftmost attainable risky portfolio.

Question 6

Which risk is generally not eliminated by holding a well-diversified portfolio?

A. Firm-specific litigation risk

B. Unsystematic management risk

C. Systematic market risk

D. Product-recall risk unique to one issuer

Correct Answer: C. Systematic market risk

Question 7

A risky portfolio has expected return 9%, standard deviation 12%, and the risk-free rate is 3%. What is its Sharpe ratio?

A. 0.25

B. 0.50

C. 0.75

D. 1.00

Correct Answer: B. 0.50

Question 8

Under the standard unconstrained model, the tangency portfolio is the risky portfolio that:

A. minimises expected return.

B. has zero covariance with every asset.

C. maximises the ex ante Sharpe ratio.

D. contains no systematic risk.

Correct Answer: C. maximises the ex ante Sharpe ratio.

Question 9

For U = E(R) − 0.5γσ², what is the effect of increasing γ while market inputs remain fixed?

A. The investor becomes less risk-averse and holds more risky assets.

B. The investor becomes more risk-averse and the optimal risky allocation decreases.

C. Portfolio variance becomes mathematically zero.

D. The tangency portfolio’s asset weights must change.

Correct Answer: B. The investor becomes more risk-averse and the optimal risky allocation decreases.

Question 10

Which practice most directly addresses unstable optimisation caused by estimation error?

A. Treating estimated returns as guaranteed

B. Ignoring covariance during crises

C. Using shrinkage, constraints and sensitivity testing

D. Maximising the number of negative weights

Correct Answer: C. Using shrinkage, constraints and sensitivity testing