Market Risk Measurement and Management

Introduction and Context

Market risk is the possibility that changes in market prices, rates, spreads, exchange rates, volatility, or correlations will reduce the value of a position or portfolio. It is distinct from credit risk, which concerns counterparty default, and liquidity risk, which concerns the ability to transact without excessive price impact. In practice, these risks interact: a market shock can increase margin requirements, impair liquidity, and raise default probabilities.

For a financial institution, market-risk management has four linked purposes: identify exposures, quantify potential loss, control risk within approved limits, and allocate capital to activities that generate adequate risk-adjusted returns. No single metric is sufficient. Sensitivity measures explain how value responds to small movements; volatility-based measures estimate distributional risk; stress tests explore severe scenarios; and performance measures assess whether compensation for risk has been adequate.

Learning Objectives

By the end of this chapter, the reader should be able to:

• measure interest-rate, equity, foreign-exchange, option, and portfolio exposures;
• calculate and interpret duration, convexity, beta, volatility, and option Greeks;
• distinguish parametric, historical-simulation, and Monte Carlo Value at Risk (VaR);
• explain Expected Shortfall (ES), backtesting, stress testing, and risk limits;
• evaluate risk-adjusted performance using measures such as the Sharpe and Treynor ratios;
• identify model, data, governance, and ethical limitations in market-risk systems.

1. Exposure Measurement: Sensitivities and Elasticities

1.1 Interest-rate risk: duration

For a fixed-income security with price P, cash flow CF_t at time t, and yield per period y,

The Macaulay duration is the present-value-weighted average time to receive the cash flows:

Duration is measured in time units. The modified duration, for yield changes expressed per compounding period, is

For a small change in yield ∆y, the first-order price approximation is

The negative sign reflects the inverse relationship between bond prices and yields. This approximation assumes a parallel, small shift in the relevant yield curve and no material change in credit spread, liquidity, or embedded-option behavior. For annualized yield changes, units must be consistent: a 25-basis-point increase is ∆y=0.0025, not 25.

Dollar duration is DD = D_mod × P, while DV01 is the approximate price change for a one-basis-point yield increase:

The sign is convention-dependent. Risk reports often display positive DV01 as the magnitude of loss from a one-basis-point rise.

1.2 Convexity

Duration is linear, but the bond-price/yield relationship is curved. Convexity measures the curvature:

For discrete cash flows, an annual-compounding approximation is

The second-order approximation is

Convexity is important for larger rate movements. For a plain option-free bond, positive convexity means that the price rises more when yields fall than it falls when yields rise by the same amount. Callable and mortgage-related securities may exhibit low or negative effective convexity because borrowers can refinance when rates fall.

1.3 Curve, spread, and basis exposures

A parallel shift is only one possible yield-curve movement. Key-rate duration measures sensitivity to a movement at a particular maturity while other key rates are held constant. It is useful for portfolios whose exposures to, for example, two-year and ten-year rates offset in aggregate duration but not locally on the curve.

Credit-sensitive instruments also have spread duration. A simplified approximation is

where D_y is yield duration, D_s is spread duration, ∆y is the risk-free yield change, and ∆s is the credit-spread change. The formula assumes separable, small changes and ignores interaction effects.

1.4 Equity and foreign-exchange exposure

For a stock or portfolio, beta measures systematic sensitivity to a benchmark return:

Here R_i is the asset return and R_m is the market return. A beta of 1.3 implies that, in the estimated sample relationship, a 1% market return has been associated with an average approximately 1.3% asset return, before considering the intercept and residual risk. Beta is not a guaranteed response and can change with leverage, business conditions, and the estimation window.

A foreign-currency position has a direct exposure to exchange-rate movements. If a domestic investor holds a foreign asset worth V_f in foreign currency and the domestic-currency value of one foreign currency unit changes by ∆s, then, approximately,

This must be adjusted for the quotation convention. Hedging with forwards or futures reduces currency exposure but introduces basis, rollover, counterparty, and collateral risks.

2. Option Greeks and Nonlinear Risk

For an option value V, the principal Greeks are local derivatives:

A local approximation is

The formula is a Taylor expansion. It is most reliable for small changes and a stable model specification. Gamma is particularly important for options because a delta-only approximation can materially understate losses during large market moves. Vega risk is relevant when implied volatility changes even if the underlying price does not.

3. Volatility and Dependence

For returns R₁,…,R_n, the sample mean is

and sample volatility is

If returns are independent and identically distributed, annualized volatility from daily volatility is approximately σ_a=σ_d√250. This scaling is an assumption, not a law: volatility clustering, autocorrelation, and non-trading-day effects can make it inaccurate.

For a portfolio with weights w, covariance matrix Σ, and return vector R,

The covariance terms show why diversification can reduce risk. Correlation is

but correlations often rise during crises, so historical diversification benefits may disappear precisely when they are most needed.

4. Value at Risk and Expected Shortfall

4.1 Definitions

At confidence level α, one-period VaR is a loss threshold not exceeded with probability α, under a specified horizon, information set, and loss convention. If L denotes loss, then

Expected Shortfall is the average loss in the tail beyond VaR:

For continuous distributions this conditional expression is direct. With discrete empirical data, ES should be calculated using a clearly specified tail-averaging rule, especially when the confidence level does not correspond to an integer observation.

VaR is easy to communicate but is not generally subadditive: diversification can appear not to reduce VaR in some distributions. ES is a coherent risk measure under standard conditions and reflects the severity of tail losses, although it is more statistically demanding to estimate.

4.2 Parametric VaR

Under a linear portfolio, normally distributed returns, and a one-period horizon, portfolio loss has standard deviation σ_p. If the mean return is μ_p, then a return-quantile formulation gives

where V is portfolio value and z_α is the standard-normal quantile. If expected return is ignored, VaR_α≈Vz_ασ_p. This is appropriate only when the sign convention, horizon, and return units are consistent. For a normal loss distribution,

where φ is the standard-normal density. The normal model may understate fat tails, skewness, and option nonlinearity.

Example. A $10 million linear portfolio has daily volatility of 1.2%, zero expected daily return, and a 99% normal VaR. Using z_0.99=2.326,

This is a one-day estimate, not a maximum possible loss and not a forecast that losses will be limited to this amount.

4.3 Historical simulation

Historical simulation applies current positions to a set of historical market changes, revalues the portfolio under each change, and takes the relevant empirical quantile. It preserves observed non-normality and dependence but assumes that the historical sample is informative about current risk. Results depend on the window, weighting scheme, treatment of missing observations, and whether full or delta revaluation is used.

4.4 Monte Carlo simulation

Monte Carlo simulation specifies stochastic models for risk factors, estimates their parameters, generates many joint scenarios, revalues the portfolio, and extracts VaR or ES from simulated losses. A simple correlated-factor construction is

where Z is a vector of independent standard-normal draws, L satisfies LLᵀ=Σ, and Σ is the factor covariance matrix. More advanced systems use stochastic volatility, jumps, regime switching, or copulas. Monte Carlo handles nonlinear instruments flexibly but is computationally expensive and model-dependent.

5. Backtesting and Model Validation

Backtesting compares realized outcomes with forecasts. For a one-day 99% VaR, approximately 1% of observations should exceed VaR if the model is correctly specified and conditions are stable. The observed exceedance rate is not expected to be exactly 1% in a finite sample.

The simplest test examines the number of exceptions using a binomial model. More advanced tests assess both unconditional coverage and independence of exceptions. Clusters of exceptions indicate that the model may fail to capture volatility dynamics. ES backtesting is harder because ES concerns the magnitude of tail losses, not merely whether a threshold was crossed; modern validation therefore uses joint VaR–ES procedures, exceedance residuals, and comparative forecasting methods.

Validation should include conceptual review, data-quality checks, benchmarking against alternative models, sensitivity analysis, parameter stability, and independent implementation testing. A model that performs well in a calm historical period may fail under a structural break.

6. Stress Testing, Limits, and Governance

Stress testing asks how a portfolio behaves under severe but plausible conditions. Historical scenarios may include a financial crisis or abrupt currency devaluation. Hypothetical scenarios can combine rate increases, equity declines, spread widening, volatility jumps, and correlation changes. Reverse stress testing begins with an unacceptable outcome and asks which combination of events could produce it.

Stress tests should be based on economic narratives as well as statistical shocks. They should identify nonlinear effects, liquidity constraints, margin calls, collateral haircuts, and management actions. A stress scenario is not assigned a routine probability merely because it is severe.

A limit framework translates risk appetite into controls. Common limits include:

• VaR and ES limits;
• gross and net notional limits;
• DV01, key-rate duration, and spread-duration limits;
• delta, gamma, and vega limits;
• stop-loss and stress-loss limits;
• concentration, basis, liquidity, and counterparty limits.

Effective governance assigns first-line ownership to trading and investment functions, independent challenge to risk management, and oversight to senior management and the board. Limit breaches require documented escalation, temporary controls, remediation, and a clear audit trail. Regulatory capital models may impose additional requirements, but regulatory compliance does not replace sound internal risk management.

7. Risk-Adjusted Performance

The Sharpe ratio is

where R_p is portfolio return, R_f is the risk-free return over the same period, and σ_p is portfolio volatility. It rewards excess return per unit of total volatility but is sensitive to non-normal returns and the measurement horizon.

The Treynor ratio is

and evaluates excess return per unit of systematic risk. It is meaningful primarily for a well-diversified portfolio. Information ratio is

where R_b is benchmark return and σ_p-b is tracking-error volatility. These measures should be interpreted with drawdown, liquidity, tail loss, turnover, leverage, and capacity information rather than used mechanically.

8. Illustrative Case: Meridian Asset Management

Background and problem

Meridian manages a $500 million bond-and-equity portfolio. Its bond book has modified duration 5.8 and an estimated DV01 of $290,000. The equity allocation has beta 1.15 to its strategic benchmark. The risk committee observes that reported 99% normal VaR is stable, while key-rate and stress results are worsening.

Analysis

A parallel 40-basis-point rise in yields implies an approximate bond loss of

or 2.32% before convexity and spread effects. The dollar approximation is also −$290,000×40=−$11.6 million, because DV01 is the one-basis-point loss magnitude. The equity book is vulnerable to a benchmark decline, and its systematic exposure is amplified by beta above one. A stress scenario combining a 40-basis-point rate increase, 8% equity decline, 100-basis-point spread widening, and a volatility jump produces a much larger loss than the normal VaR estimate because the portfolio includes callable bonds and options.

Approach and solution

Meridian adopts a layered framework: daily parametric VaR for rapid monitoring; historical simulation using crisis observations; full-revaluation Monte Carlo for options; key-rate duration and spread-duration limits; and monthly reverse stress testing. It supplements the 99% VaR limit with ES and scenario-loss limits. The firm also validates whether its historical window contains materially different volatility regimes and requires independent review of pricing and risk-factor mappings.

Outcome and learning points

The committee reduces concentrated long-duration exposure, adds a partial hedge using interest-rate futures, and limits short-volatility positions. The hedge is monitored for basis and roll risk rather than assumed to be perfect. The central lesson is that a stable aggregate VaR can conceal changing factor concentrations, nonlinear exposures, and tail dependence. Risk measurement is most useful when metrics lead to specific, governed decisions.

9. Limitations, Ethics, and Contemporary Practice

Risk models can create false precision. Parameter uncertainty, data snooping, stale prices, survivorship bias, unobserved liquidity costs, and inappropriate aggregation can all distort results. A model may be technically correct but operationally unsafe if users misunderstand its assumptions.

Ethically, institutions must avoid presenting VaR as a guarantee, suppressing unfavorable scenarios, or using model complexity to obscure risk. Governance should require transparent documentation, model inventories, version control, outcome monitoring, and escalation of known weaknesses. AI and machine-learning methods can improve volatility forecasting, anomaly detection, and scenario generation, but they may also introduce unstable relationships, opaque decisions, leakage from future information, and discriminatory or poorly governed outputs. Explainability, challenger models, stress testing, and human accountability remain essential.

Contemporary practice increasingly combines traditional factor models with high-frequency data, regime-sensitive volatility models, expected shortfall, climate and transition scenarios, and intraday risk monitoring. These developments improve responsiveness but do not eliminate the need for economic judgment. The most robust architecture is pluralistic: several complementary measures, independently challenged and linked to decisions.

Chapter Summary

Market risk management begins with identifying risk-factor exposures. Duration and DV01 measure first-order interest-rate sensitivity, while convexity improves estimates for larger yield changes. Key-rate and spread measures reveal curve and credit-spread concentrations. Beta measures systematic equity exposure; portfolio volatility depends on the covariance matrix; and Greeks describe the local sensitivity and nonlinearity of options.

VaR summarizes a loss quantile, whereas ES describes the average loss in the tail. Parametric methods are efficient but assumption-dependent, historical simulation is empirically grounded but regime-sensitive, and Monte Carlo is flexible but model-intensive. Backtesting evaluates forecast behavior, while stress and reverse stress testing examine severe outcomes that ordinary probability models may miss. Limits, escalation, independent validation, and board-level risk appetite convert measurement into management. Finally, Sharpe, Treynor, and information ratios support—but do not replace—judgment about liquidity, drawdown, leverage, capacity, and tail risk.

Key Terms

Market risk: Risk of loss from changes in market prices or rates.

Macaulay duration: Present-value-weighted average timing of fixed-income cash flows.

Modified duration: First-order percentage price sensitivity to yield.

DV01: Approximate price change for a one-basis-point yield movement.

Convexity: Second-order curvature of the price–yield relationship.

Key-rate duration: Sensitivity to a specific maturity-point yield change.

Spread duration: Sensitivity to a credit-spread movement.

Beta: Covariance with a benchmark divided by benchmark variance.

Greek: Derivative-based sensitivity of an option value to a risk factor.

Volatility: Dispersion of returns, commonly measured by standard deviation.

Covariance matrix: Matrix describing variances and co-movements of risk factors.

Value at Risk: Specified quantile of a loss distribution.

Expected Shortfall: Average loss beyond a specified VaR threshold.

Parametric VaR: VaR based on an assumed distribution and estimated parameters.

Historical simulation: VaR based on reapplying observed historical changes.

Monte Carlo simulation: VaR based on generated scenarios from a stochastic model.

Backtesting: Comparing forecasts with subsequent realized outcomes.

Stress testing: Evaluating portfolio performance under severe scenarios.

Risk limit: Approved maximum exposure or loss measure.

Risk-adjusted performance: Return evaluated relative to risk taken.

Review Questions

Short Answer Questions

1. Why is modified duration usually more useful than Macaulay duration for estimating a bond’s percentage price change?
2. What additional information does convexity provide beyond duration?
3. Distinguish VaR from Expected Shortfall.
4. Why can historical diversification benefits fail during a crisis?
5. What is the purpose of backtesting a VaR model?

Long Answer Questions

1. Compare parametric, historical-simulation, and Monte Carlo VaR in terms of assumptions, implementation, and model risk.
2. Explain how duration, key-rate duration, spread duration, and convexity can be integrated into a fixed-income risk system.
3. Discuss the advantages and limitations of Greeks for managing an options portfolio.
4. Design a governance framework for stress testing, limit breaches, and independent model validation.
5. Critically evaluate the use of Sharpe, Treynor, and information ratios in assessing investment performance.

Analytical / Application Questions

1. A bond portfolio worth $80 million has modified duration 4.5. Estimate the percentage and dollar price change from a 35-basis-point yield increase, ignoring convexity. State the assumptions.
2. A portfolio has daily volatility of 1.5%, value $20 million, zero expected return, and normally distributed returns. Estimate one-day 95% VaR using z_0.95=1.645, then explain why the estimate may understate option-portfolio risk.
3. Construct a stress scenario for a leveraged multi-asset portfolio involving rates, equities, spreads, volatility, and liquidity. Identify at least three limits or management actions that should be triggered.

Multiple Choice Questions with Answers

Question 1

A bond has modified duration 6.0. Ignoring convexity, what is the approximate percentage price change for a 25-basis-point yield increase?

A. −0.15%

B. −1.50%

C. +0.15%

D. +1.50%

Correct Answer: A. −0.15%

Question 2

Which measure focuses on the average loss beyond a specified loss quantile?

A. Beta

B. DV01

C. Expected Shortfall

D. Macaulay duration

Correct Answer: C. Expected Shortfall

Question 3

Under the normal linear VaR model, a portfolio’s risk is primarily determined by its value, volatility, and which additional quantity?

A. The number of securities

B. The relevant distribution quantile

C. The accounting depreciation rate

D. The coupon frequency alone

Correct Answer: B. The relevant distribution quantile

Question 4

Which Greek measures the change in delta as the underlying asset price changes?

A. Vega

B. Rho

C. Theta

D. Gamma

Correct Answer: D. Gamma

Question 5

A major strength of historical simulation is that it can preserve:

A. A guaranteed normal distribution

B. Observed non-normality and dependence

C. Perfect future predictability

D. Zero model risk

Correct Answer: B. Observed non-normality and dependence

Question 6

If a portfolio has $10 million value and daily volatility of 1.2%, with zero mean and a 99% normal quantile of 2.326, its approximate one-day VaR is:

A. $27,912

B. $120,000

C. $279,120

D. $2,326,000

Correct Answer: C. $279,120

Question 7

What does a cluster of VaR exceptions usually suggest?

A. The model may not capture changing volatility dynamics

B. The model is necessarily conservative

C. The portfolio has no market risk

D. The confidence level is exactly correct

Correct Answer: A. The model may not capture changing volatility dynamics

Question 8

Which measure is most directly associated with systematic equity-market exposure?

A. Beta

B. Convexity

C. DV01

D. Vega

Correct Answer: A. Beta

Question 9

A key-rate duration framework is especially useful for identifying:

A. Specific maturity-point yield-curve exposures

B. Only foreign-exchange exposures

C. Accounting profits

D. Counterparty legal capacity

Correct Answer: A. Specific maturity-point yield-curve exposures

Question 10

Which statement best describes Monte Carlo risk measurement?

A. It uses only the single worst historical observation

B. It generates scenarios from specified stochastic models

C. It eliminates calibration risk

D. It is always faster than parametric methods

Correct Answer: B. It generates scenarios from specified stochastic models