Modern Portfolio Theory and Diversification

Introduction

Investors rarely evaluate securities in isolation. A bank treasury, mutual fund, insurer or pension fund must determine how individual assets interact within a portfolio. An asset that appears risky by itself may nevertheless reduce total portfolio risk when its returns do not move closely with those of existing holdings.

Modern Portfolio Theory (MPT), developed principally by Harry Markowitz, formalizes this insight through mean–variance analysis. Expected return represents reward, while variance or standard deviation represents risk. Portfolio construction then becomes a problem of selecting combinations that offer the highest expected return for a given variance—or the lowest variance for a given expected return.

MPT provides the foundations for diversification, efficient frontiers, utility-based portfolio choice, capital allocation and the separation principle. Its practical use nevertheless requires caution because expected returns and covariances are estimated rather than known.

Learning Objectives

After studying this chapter, students should be able to:

1. Calculate portfolio expected return, variance and standard deviation.
2. Explain how correlation determines diversification benefits.
3. Derive and interpret minimum-variance and efficient portfolios.
4. Use risk-aversion parameters and indifference curves in portfolio choice.
5. identify the tangency portfolio and apply the separation principle.
6. Evaluate diversification across assets, sectors and countries.
7. Recognize estimation error, implementation constraints and model limitations.
8. Apply MPT responsibly in banking, insurance and investment practice.

1. Return and Risk in a Portfolio

1.1 Portfolio return

For a portfolio containing n assets, the realized portfolio return is the weighted average of individual asset returns:

Here, R_p is the portfolio return, w_i is the proportion invested in asset i, and R_i is its return. For a fully invested portfolio, \sum w_i=1. Negative weights represent short positions when short selling is permitted.

Taking expectations gives:

Returns and expected returns must refer to the same period and should be expressed consistently as decimals or percentages.

1.2 Portfolio variance

Portfolio risk depends on both individual asset risks and their co-movements:

In this expression, σ_p² is portfolio variance, \boldsymbol{Σ} is the covariance matrix and \mathbf{w} is the weight vector. If returns are measured annually, covariance and variance are in squared annual-return units; standard deviation is in annual-return units.

For two assets, A and B:

Here, ρ_AB is the correlation between the two returns, where -1≤ρ_AB≤1. Covariance is related to correlation by Cov(R_A,R_B)=ρ_ABσ_Aσ_B.

Worked Example 1: A two-asset portfolio

Suppose Asset A has an expected annual return of 10% and standard deviation of 15%. Asset B has an expected return of 6% and standard deviation of 8%. Their correlation is 0.25. The portfolio invests 60% in A and 40% in B.

Expected return is:

Variance is:

Therefore:

The weighted average of the two standard deviations is 12.2%, but portfolio standard deviation is only 10.28%. The reduction results from imperfect correlation.

2. The Logic of Diversification

Diversification reduces risk when assets are not perfectly positively correlated. Its effectiveness depends on covariance—not simply on the number of securities held.

• If ρ=+1, the assets move proportionately in the same direction, and there is no correlation-based risk reduction.
• If -1<ρ<+1, diversification reduces portfolio variance relative to perfect positive correlation.
• If ρ=-1, an appropriately weighted two-asset portfolio may have zero variance, although stable perfect negative correlation is rare in financial markets.

Adding many securities primarily removes idiosyncratic risk, which is specific to issuers or projects. Market-wide systematic risk cannot normally be eliminated by holding more securities from the same risk system.

Under a one-factor model:

and, assuming Cov(R_M,ε_i)=0,

The first component is systematic market risk; the second is idiosyncratic risk. In a well-diversified portfolio, independent or weakly correlated residual risks tend to offset one another.

Diversification can fail during crises because correlations are not constant. Assets that appeared weakly correlated in normal markets may fall simultaneously when liquidity disappears or investors deleverage.

3. The Feasible Set and Efficient Frontier

The feasible set contains every risk–return combination attainable from the available assets under specified constraints. Its shape depends on expected returns, covariances and restrictions such as no short selling.

The global minimum-variance portfolio (GMVP) is the fully invested portfolio with the lowest possible variance. The upper boundary extending from the GMVP is the efficient frontier. Portfolios below this boundary are inefficient because another feasible portfolio offers either:

• a higher expected return for the same risk; or
• lower risk for the same expected return.

Figure: The efficient frontier is exactly the upper boundary of the shaded feasible set above the global minimum-variance portfolio.

3.1 Global minimum-variance weights

Suppose short selling is allowed, the portfolio is fully invested, and the covariance matrix \boldsymbol{Σ} is positive definite. Minimizing variance subject to \mathbf{1}'\mathbf{w}=1 gives:

Here, \mathbf{1} is a column vector of ones. Positive definiteness ensures a unique minimum in the unconstrained problem. If short selling or position sizes are constrained, quadratic programming is generally required and the formula need not apply.

3.2 Derivation of the mean–variance frontier

To find the lowest-variance portfolio for a target expected return m, solve:

subject to:

Define:

If \boldsymbol{Σ} is positive definite and expected returns are not identical across all assets, then D>0. The minimum-variance weights for target return m are:

The corresponding frontier variance is:

The upper branch, where expected return is at least the GMVP expected return, is efficient. The lower branch is minimum variance for its target return but is mean–variance inefficient relative to portfolios on the upper branch.

4. Investor Preferences, Utility and Indifference Curves

The efficient frontier identifies attractive opportunities but does not identify a single portfolio for every investor. Selection also depends on risk preferences.

A common approximation to investor utility is:

Here, U is mean–variance utility, γ>0 is the coefficient of risk aversion, expected return is measured per period, and variance is measured for the same period. A larger γ indicates greater aversion to variance.

For a fixed utility level \bar U, the indifference curve is:

In expected-return–standard-deviation space, this curve is upward sloping and convex. Investors require progressively more expected return to accept additional risk. The preferred risky portfolio is found where the highest attainable indifference curve is tangent to the efficient frontier, provided the portfolio choice is unconstrained and the utility representation is appropriate.

Mean–variance utility is exactly consistent with expected utility when returns are normally distributed or utility is quadratic over the relevant range. Outside these conditions, it is often used as a local approximation and may not capture skewness, tail risk or loss aversion.

5. Introducing a Risk-Free Asset

A risk-free asset has a known return R_f over the investment horizon and zero return variance in the investor's reference currency. A complete portfolio allocating proportion y to risky portfolio P and 1-y to the risk-free asset has:

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

The slope is portfolio P's Sharpe ratio. If borrowing and lending occur at the same risk-free rate and short selling is unrestricted, the CAL with the greatest positive slope is tangent to the risky-asset frontier.

5.1 Tangency portfolio

Let the excess-return vector be:

With unrestricted risky-asset short selling and a positive-definite covariance matrix, define:

If \mathbf{e}\ne\mathbf{0} and d>0, the fully invested risky portfolio that maximizes the positive Sharpe ratio is:

The condition d>0 is material. A merely nonzero denominator is insufficient. If d<0, the normalized stationary portfolio has a negative Sharpe ratio and is not the positive maximum-Sharpe portfolio. If d=0, the optimal positive ray cannot be normalized into a finite fully invested risky portfolio. With no-short-sale, leverage or other constraints, the tangency portfolio must instead be obtained through constrained optimization.

Worked Example 2: Tangency weights

Use the two assets from Worked Example 1 and let R_f=3\%. Their covariance matrix and excess-return vector are:

Calculation gives:

Thus:

The tangency portfolio has:

Its Sharpe ratio is approximately:

5.2 Optimal allocation between risky and risk-free assets

For quadratic mean–variance utility, unrestricted allocation to the tangency portfolio is:

This assumes the investor can borrow or lend at R_f and that utility is based only on mean and variance. Using γ=8 in the example:

The investor allocates approximately 76.2% to the tangency portfolio and 23.8% to the risk-free asset. The complete portfolio has expected return of approximately 6.61% and standard deviation of approximately 6.72%.

6. The Separation Principle

The separation principle divides portfolio choice into two decisions:

1. Determine the optimal risky portfolio using market expectations and covariances.
2. Combine that risky portfolio with the risk-free asset according to the investor's risk aversion.

Under the classical assumptions, all investors facing the same inputs and risk-free rate hold the same tangency portfolio but choose different values of y. Conservative investors lend at the risk-free rate; aggressive investors may invest all wealth in the tangency portfolio or borrow to leverage it.

This result depends on strong assumptions: homogeneous expectations, frictionless markets, unrestricted borrowing and short selling, a common risk-free rate, and mean–variance preferences. In practice, taxes, liabilities, regulation, borrowing spreads and investment constraints cause investors to hold different risky portfolios.

7. Diversification Across Assets, Sectors and Countries

7.1 Asset-class diversification

Combining equities, government bonds, corporate credit, property and cash may reduce dependence on a single economic risk driver. Asset labels alone are insufficient: corporate bonds and equities, for example, may both suffer when corporate solvency deteriorates.

7.2 Sector diversification

Sector diversification limits exposure to industry-specific shocks such as commodity-price changes, technological disruption, regulation or litigation. Superficial diversification may occur when a portfolio owns many securities concentrated in the same sector or economic factor.

Risk contribution provides a better diagnostic than capital weights. The marginal contribution of asset i to portfolio variance is:

An asset with a small weight can make a large contribution to risk if it is volatile and strongly correlated with the rest of the portfolio.

7.3 International diversification

International assets may provide access to different business cycles, industries and monetary regimes. However, domestic investors also face currency risk. If R_L is the local-currency asset return and R_FX is the return from the foreign currency relative to the domestic currency, the exact domestic-currency return is:

or:

The approximation R_D≈R_L+R_FX is reasonable only when returns are sufficiently small. International investing also introduces taxation, capital controls, custody, political risk, accounting differences and market-access constraints.

8. Illustrative Case: Diversifying an Insurance Investment Fund

Background

This fictional case concerns Meridian General Insurance. Its investment portfolio must support claim payments while generating a competitive return. Management is considering domestic bonds, domestic equities and international equities. Annual estimates are:

Estimated correlations are 0.10 between bonds and domestic equities, 0.05 between bonds and international equities, and 0.65 between the two equity categories.

Problem

The existing 50% bond–50% domestic equity portfolio is geographically concentrated. Management wants to assess a candidate allocation of 40% bonds, 35% domestic equities and 25% international equities.

Analysis

The candidate portfolio's expected return is:

Its variance is:

The existing domestic portfolio has expected return 7.50%, variance 0.00954 and standard deviation approximately 9.77%. With a 3% risk-free rate, its Sharpe ratio is approximately 0.461, compared with approximately 0.479 for the candidate portfolio.

Approach and Solution

The candidate has slightly greater volatility but a higher expected return and estimated Sharpe ratio. Meridian should not adopt it mechanically. As an insurer, it should also:

• model claim-liability duration and liquidity needs;
• stress equity and currency losses jointly;
• use covariance shrinkage rather than raw sample covariances alone;
• impose country, sector and counterparty limits;
• assess capital and solvency requirements; and
• determine whether foreign-currency exposure should be hedged.

Outcome

Management adopts the diversified allocation as a strategic benchmark, subject to currency hedging and quarterly risk monitoring. It also tests a severe scenario in which both equity markets decline and their correlation rises sharply.

Learning Points

Diversification should be evaluated through covariance and total portfolio risk, not asset count. A statistically efficient allocation may still be unsuitable if it conflicts with liabilities, liquidity obligations or regulatory capital requirements.

9. Estimation Error and Model Risk

MPT treats expected returns and covariances as inputs, but these quantities are unknown. They must be estimated from historical data, market prices or forecasts.

Expected returns are especially difficult to estimate. If independent returns have annual standard deviation σ and there are T annual observations, the standard error of the sample mean is approximately:

For σ=20\% and T=10, the standard error is about 6.32%, which is large relative to typical differences in expected returns. Serial dependence and changing regimes may make the simple formula inappropriate.

Optimization can amplify small input errors and produce unstable, concentrated weights. Common responses include:

• Shrinkage: combining sample estimates with structured targets.
• Bayesian methods: blending prior beliefs with observed information.
• Black–Litterman analysis: integrating equilibrium returns with investor views.
• Robust optimization: selecting portfolios that perform acceptably across ranges of inputs.
• Weight constraints: limiting short positions, leverage and concentration.
• Resampling and simulation: examining sensitivity across plausible datasets.
• Stress testing: changing correlations, volatility, liquidity and exchange rates under adverse scenarios.

Backtests must avoid look-ahead bias, survivorship bias, data snooping and unrealistic assumptions about costs. Machine-learning forecasts can identify nonlinear patterns, but they may intensify overfitting and do not remove the need for economic justification and out-of-sample validation.

10. Limitations of Mean–Variance Analysis

Important limitations include:

1. Variance treats upside and downside equally. Investors may care more about losses, drawdowns or shortfall risk.
2. Returns may be non-normal. Skewness, fat tails and discontinuous losses can matter materially.
3. Inputs are unstable. Expected returns, volatility and correlations change over time.
4. Single-period focus. Classical MPT does not directly model multiperiod consumption, trading or liability dynamics.
5. Market frictions are omitted. Taxes, transaction costs, illiquidity and borrowing spreads affect feasible portfolios.
6. Optimization may produce extreme weights. Mathematically efficient portfolios can be operationally fragile.
7. Correlation is not causation or structural stability. Historical relationships may fail during crises.

Alternative or complementary measures include Value at Risk, Expected Shortfall, downside deviation, maximum drawdown, scenario analysis and asset–liability modeling.

11. Governance and Professional Practice

Financial institutions should treat portfolio optimization as a governed decision process rather than an automatic trading rule. Good practice includes documented assumptions, independent validation, data-quality controls, benchmark selection, limit monitoring and approval procedures for model changes.

Portfolio managers owe duties of competence, care and fair treatment. They should not present estimated efficient frontiers as guaranteed outcomes. Environmental, social and governance information may affect expected cash flows and risk, but inconsistent data and subjective ratings require transparent methodology. Client suitability, insurance solvency rules, liquidity requirements and fiduciary obligations take precedence over a model's unconstrained mathematical optimum.

Chapter Summary

Modern Portfolio Theory evaluates assets through their contribution to total portfolio return and variance. Portfolio expected return is a weighted average, whereas portfolio variance depends on covariance and correlation. Imperfect correlation creates diversification benefits and can reduce idiosyncratic risk.

The feasible set contains all attainable portfolios, while the efficient frontier contains portfolios offering the maximum expected return for each risk level. Investor risk aversion and indifference curves determine the preferred position. With a risk-free asset, the positive maximum-Sharpe tangency portfolio defines the best capital allocation line under classical assumptions. The separation principle distinguishes selection of the risky portfolio from the investor's financing decision.

Diversification should extend across securities, sectors, factors, countries and currencies, while remaining consistent with liabilities and constraints. Because optimization is highly sensitive to estimated inputs, professional implementation requires shrinkage, robust analysis, stress testing, governance and transparent recognition of model limitations.

Key Terms

Mean–variance analysis: Evaluation of portfolios using expected return and return variance.

Covariance: Measure of how two asset returns vary together.

Correlation: Standardized covariance bounded between −1 and +1.

Diversification: Combining imperfectly correlated exposures to control portfolio risk.

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

Idiosyncratic risk: Asset- or issuer-specific risk that can often be diversified.

Feasible set: All portfolios attainable under specified inputs and constraints.

Efficient frontier: Portfolios offering the greatest expected return for each risk level.

Global minimum-variance portfolio: Fully invested portfolio having the lowest attainable variance.

Risk aversion: Preference for lower risk when expected return is otherwise equal.

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

Risk-free asset: Asset with a known return over the relevant horizon in the reference currency.

Capital allocation line: Risk–return combinations formed from a risky portfolio and a risk-free asset.

Sharpe ratio: Expected excess return per unit of return standard deviation.

Tangency portfolio: Risky portfolio generating the highest attainable positive Sharpe ratio under applicable assumptions.

Separation principle: Division of portfolio choice into risky-portfolio selection and risk-free allocation.

Estimation error: Difference between estimated model inputs and their unknown true values.

Shrinkage: Combining noisy estimates with a structured target to improve stability.

Robust optimization: Portfolio selection allowing explicitly for uncertainty in model inputs.

Model risk: Risk of loss or poor decisions arising from flawed models, assumptions or implementation.

Review Questions

Short Answer Questions

1. Why does portfolio variance depend on covariance as well as individual asset variance?
2. What distinguishes an efficient portfolio from an inefficient portfolio?
3. Explain the difference between systematic and idiosyncratic risk.
4. What does the risk-aversion coefficient represent in mean–variance utility?
5. Why can international diversification introduce additional currency risk?

Long Answer Questions

1. Derive the expected return and variance of a two-asset portfolio and explain the role of correlation.
2. Explain the construction and economic meaning of the feasible set, GMVP and efficient frontier.
3. Discuss the tangency portfolio and state the conditions required for its closed-form weight formula to maximize the positive Sharpe ratio.
4. Critically evaluate the separation principle and its practical limitations.
5. Discuss major sources of estimation error in MPT and the techniques used to manage them.

Analytical / Application Questions

1. Assets X and Y have standard deviations of 12% and 20%, correlation 0.30, and portfolio weights of 70% and 30%. Calculate portfolio standard deviation.
2. A risky portfolio offers an expected return of 9%, standard deviation of 12%, and the risk-free rate is 3%. For an investor with γ=5, calculate the optimal risky allocation under unrestricted mean–variance assumptions.
3. Design a diversification and model-governance process for an insurer whose current portfolio is concentrated in domestic bank equities and government bonds.

Multiple Choice Questions with Answers

Question 1

A portfolio invests 60% in an asset returning 10% and 40% in an asset returning 6%. What is its expected return?

A. 6.8%

B. 7.6%

C. 8.4%

D. 9.2%

Correct Answer: C. 8.4%

Question 2

For the portfolio in Worked Example 1, what is the calculated annual variance?

A. 0.010564

B. 0.012200

C. 0.102800

D. 0.014400

Correct Answer: A. 0.010564

Question 3

Which statement best defines an efficient portfolio?

A. It has the smallest number of securities.

B. It provides the highest expected return for its level of risk.

C. It contains no risky securities.

D. It always has the highest expected return available.

Correct Answer: B. It provides the highest expected return for its level of risk.

Question 4

If two risky assets have correlation below +1, what is generally true?

A. Combining them can provide diversification benefits.

B. Their portfolio variance must be zero.

C. Their expected returns must be equal.

D. Both assets must have negative returns together.

Correct Answer: A. Combining them can provide diversification benefits.

Question 5

In U=E(R_p)-(γ/2)σ_p², an increase in γ indicates:

A. lower portfolio variance by definition

B. a higher risk-free rate

C. greater investor risk aversion

D. greater expected portfolio return

Correct Answer: C. greater investor risk aversion

Question 6

Under unrestricted short selling, which additional sign condition is required for the standard normalized tangency-weight formula to maximize the positive Sharpe ratio?

A. \mathbf{1}'\boldsymbol{Σ}^-1(\boldsymbol{μ}-R_f\mathbf{1})>0

B. \mathbf{1}'\boldsymbol{Σ}^-1(\boldsymbol{μ}-R_f\mathbf{1})<0

C. \det(\boldsymbol{Σ})<0

D. R_f>\max(μ_i)

Correct Answer: A. \mathbf{1}'\boldsymbol{Σ}^-1(\boldsymbol{μ}-R_f\mathbf{1})>0

Question 7

What does the classical separation principle imply?

A. Every investor must hold only the risk-free asset.

B. Risky-portfolio selection is separated from the investor's allocation between risky and risk-free assets.

C. International and domestic assets must be analyzed separately.

D. Systematic risk can be separated completely from return.

Correct Answer: B. Risky-portfolio selection is separated from the investor's allocation between risky and risk-free assets.

Question 8

In the illustrative insurance case, what is the candidate portfolio's expected annual return?

A. 7.25%

B. 7.50%

C. 7.75%

D. 8.25%

Correct Answer: C. 7.75%

Question 9

If annual return standard deviation is 20% and there are 10 independent annual observations, what is the approximate standard error of the sample mean?

A. 2.00%

B. 4.47%

C. 6.32%

D. 20.00%

Correct Answer: C. 6.32%

Question 10

Which formula gives the exact domestic-currency return on an unhedged foreign investment?

A. R_D=R_L-R_FX

B. R_D=R_L+R_FX in all circumstances

C. R_D=(R_L)(R_FX)

D. R_D=R_L+R_FX+R_LR_FX

Correct Answer: D. R_D=R_L+R_FX+R_LR_FX