Market Risk and Trading Book Management

Introduction and Context

Market risk is the possibility that changes in market prices, rates, spreads, or volatilities will reduce the value of a bank’s positions or impair its earnings. In a banking institution, market risk arises in both the banking book and the trading book. The banking book primarily contains positions held to collect contractual cash flows or for longer-term relationship purposes; the trading book contains instruments held with trading intent, market-making intent, or to profit from short-term price movements.

Trading-book management is therefore more than calculating a daily profit-and-loss figure. It requires a coherent system for identifying risk factors, measuring sensitivities, aggregating exposures, setting limits, validating models, executing hedges, conducting stress tests, and governing trader behaviour. Modern regulation, particularly the Basel market-risk framework, has strengthened the distinction between trading and banking books while recognising that risks cannot be controlled by a single metric.

This chapter develops the principal tools used to measure interest-rate, foreign-exchange, equity, commodity, basis, and volatility risk. It then examines Value at Risk (VaR), Expected Shortfall (ES), backtesting, stress testing, hedging, trading-book regulation, and model limitations.

Learning Objectives

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

1. Distinguish trading-book market risk from banking-book interest-rate risk.
2. Explain and calculate major risk sensitivities, including delta, duration, modified duration, convexity, vega, and basis exposure.
3. Interpret VaR and ES, including their assumptions and limitations.
4. Design a practical framework for backtesting and stress testing.
5. Evaluate hedging strategies for rate, currency, equity, commodity, basis, and volatility risks.
6. Explain the logic of contemporary trading-book regulation and the role of risk governance.
7. Critically assess model risk, liquidity risk, data limitations, and behavioural incentives.

1. The Architecture of Trading-Book Risk

A trading position has value that depends on one or more observable or inferred risk factors. Examples include a yield-curve point, an exchange rate, an equity index, a commodity future, a credit spread, an implied volatility, or a correlation parameter. A useful conceptual representation is:

where V is portfolio value, xᵢ are market risk factors, and t is time. Market risk is generated when a change in any xᵢ changes V. The risk-management task is to estimate the distribution of the change in value over a specified horizon and confidence level, subject to realistic assumptions about trading, liquidity, and hedging.

A trading-book risk system normally contains five connected layers:

The distinction between risk measurement and risk control is important. A precise measure does not itself reduce risk. Risk is reduced through limits, diversification, hedging, position reduction, collateral, capital allocation, and disciplined escalation.

2. Risk-Factor Taxonomy

2.1 Interest-rate risk

Interest-rate risk affects bonds, swaps, repos, options, mortgages, and many structured products. It may arise from parallel shifts in the yield curve, changes in curve shape, movements in forward rates, or differences between reference rates. A portfolio can be hedged against a parallel shift while remaining highly exposed to steepening, flattening, or curvature changes.

2.2 Foreign-exchange risk

Foreign-exchange (FX) risk results from changes in spot exchange rates, forward points, cross-currency basis, and foreign interest rates. A bank with assets denominated in one currency and funding in another has structural exposure; a trading desk may have directional, option, or correlation exposure.

2.3 Equity risk

Equity risk includes movements in individual shares, equity indices, dividends, correlations, and implied volatility. A long call option, for example, can lose value even when the underlying share is unchanged if implied volatility falls or time decay accelerates.

2.4 Commodity risk

Commodity exposures include spot prices, futures curves, location spreads, quality spreads, seasonality, and storage costs. Oil, gas, metals, and agricultural products can exhibit abrupt jumps and limited liquidity. A hedge using a nearby futures contract may leave material basis risk.

2.5 Basis risk

Basis risk is the risk that two prices or rates expected to move together do not do so. Examples include three-month reference-rate exposure hedged with an overnight-index swap, a corporate bond hedged with a government bond, and jet-fuel exposure hedged with crude oil. Basis is often small in ordinary markets but can become dominant during stress.

2.6 Volatility and correlation risk

Option values depend not only on the underlying price but also on implied volatility, the volatility surface, interest rates, dividends, correlation, and time. Volatility risk is commonly summarised by vega; changes in the slope or curvature of the volatility surface create additional risks not captured by a single vega number.

3. Local Sensitivities and Scenario Analysis

For a small change in risk factors, a first-order approximation is:

Here ∆V is the approximate change in value, ∂V/∂xᵢ is the sensitivity to factor xᵢ, and ∆xᵢ is the factor change. The approximation is most reliable for small movements and reasonably linear instruments.

A second-order approximation incorporates curvature:

The second derivative terms capture convexity and cross-gamma. They are particularly important for options, mortgage-related instruments, and portfolios exposed to large market movements.

3.1 Interest-rate duration

For a fixed-rate bond with price P and yield y, modified duration is:

D_mod is measured in years when y is expressed as a decimal change. For a small yield change ∆y:

The negative sign indicates that bond prices generally fall when yields rise. The formula assumes a sufficiently small, approximately parallel yield movement and a fixed cash-flow structure.

For a bond paying coupons m times annually, Macaulay duration and modified duration are related by:

where y is the annualised yield and D_Mac is Macaulay duration. The relation assumes conventional yield-to-maturity discounting and consistent compounding.

Convexity improves the approximation:

where C is price convexity, defined as C = (1/P)(∂²P/∂y²). Convexity is expressed in squared years under common conventions. For a plain fixed-rate bond, positive convexity means that the price gain from a fall in yield is usually larger than the price loss from an equal rise in yield.

#### Worked example: duration and convexity

Suppose a bond is priced at 98.00, has modified duration of 4.2 years, and convexity of 28. A yield increase of 50 basis points means ∆y = 0.005. Then:

∆P/P ≈ −4.2(0.005) + ½(28)(0.005²)

∆P/P ≈ −0.021 + 0.00035 = −0.02065

The estimated price decline is 2.065%, or approximately 2.02 currency units per 98.00 of price. The estimate is local; for a large movement, the full pricing model or revaluation should be used.

3.2 Key-rate duration

Key-rate duration measures sensitivity to a specific maturity point on the yield curve, such as two, five, ten, or thirty years. It is superior to a single duration measure when curve shape changes matter. A portfolio may have near-zero total duration but significant offsetting two-year and ten-year key-rate exposures.

3.3 FX delta and forward exposure

For a position whose domestic-currency value is V(S), where S is the exchange rate, FX delta is:

The unit depends on how S is quoted. A disciplined risk system records both the quotation convention and the currency of the sensitivity. FX options also possess gamma and vega, while forward contracts have exposure to spot, interest-rate differentials, and cross-currency basis.

3.4 Equity and commodity sensitivities

Equity delta measures the change in portfolio value for a unit change in the equity price. Commodity delta can be reported per unit of the physical commodity or per futures-price movement. Futures-curve exposures should be mapped by contract month because a position in a near-month contract is not identical to a position in a deferred contract.

3.5 Option sensitivities

Common option Greeks include:

• Delta: first-order sensitivity to the underlying price.
• Gamma: rate of change of delta; a second-order price sensitivity.
• Vega: sensitivity to implied volatility, usually for a one-percentage-point volatility change.
• Theta: sensitivity to the passage of time, holding other variables constant.
• Rho: sensitivity to interest rates.

Greeks are model-dependent. For American, path-dependent, or illiquid options, numerical sensitivities may depend strongly on the pricing model, calibration, bump size, and treatment of discrete events.

4. Value at Risk and Expected Shortfall

4.1 Value at Risk

VaR at confidence level α over horizon h is the loss threshold that is exceeded with probability approximately 1 − α under the specified model and information set. If Lₕ is the loss over horizon h, VaR satisfies:

Equivalently, VaR is a quantile of the loss distribution. A 99% one-day VaR of 10 million means that, under the model, losses greater than 10 million are expected on about 1% of comparable days. It does not mean that the maximum loss is 10 million.

Common methods are:

1. Historical simulation: applies observed historical changes to current positions. It is transparent and preserves non-linear revaluation if full pricing is used, but depends on the chosen window and may omit unprecedented events.
2. Parametric VaR: assumes a distribution, often normality, and uses exposures and a covariance matrix. It is fast but can be poor for options, fat tails, and unstable correlations.
3. Monte Carlo VaR: simulates risk factors from a specified stochastic model and revalues the portfolio. It is flexible but computationally intensive and model-sensitive.

Under simple normal, linear, one-day assumptions, a portfolio VaR may be expressed as:

where z_α is the standard-normal quantile, σ_P is portfolio return volatility, and V is portfolio value or exposure base. This equation is not generally valid for nonlinear portfolios or non-normal returns.

4.2 Expected Shortfall

Expected Shortfall at level α is the expected loss conditional on losses being at or beyond the VaR threshold:

For continuous distributions this expression is straightforward. For discrete empirical distributions, the tail may require a weighted quantile definition. ES gives information about tail severity, not merely the boundary of the tail. It is generally a coherent risk measure under appropriate conditions, including subadditivity, whereas VaR can fail subadditivity in some distributions.

The selection of confidence level and horizon is a governance decision. Longer horizons require assumptions about position changes, liquidity, funding, margin, and the ability to hedge. Square-root-of-time scaling, VaR_T ≈ VaR₁√T, assumes independently and identically distributed returns and is unreliable when returns are autocorrelated, volatile, or subject to liquidity effects.

5. Backtesting, P&L Attribution, and Model Validation

Backtesting compares predicted risk with subsequent realised outcomes. For a one-day VaR model, an exception occurs when the realised loss exceeds the previous day’s VaR. If the model is correctly calibrated at 99%, approximately 1% of observations should be exceptions, although random variation is expected.

A useful backtest examines:

• Unconditional coverage: whether the number of exceptions is plausible.
• Independence: whether exceptions cluster rather than occur randomly.
• Conditional coverage: whether both frequency and independence are acceptable.
• P&L attribution: whether hypothetical or risk-theoretical P&L is sufficiently aligned with actual P&L.

A model can pass a simple exception-count test while failing economically because losses cluster during crises. Conversely, a small sample can produce apparent failures by chance. Backtesting therefore supports, but does not replace, judgement and stress testing.

P&L attribution compares risk-theoretical P&L, generated from modelled risk factors, with hypothetical or actual P&L. Large unexplained differences may indicate missing risk factors, incorrect mappings, stale prices, valuation adjustments, intraday activity, or booking problems.

Model validation should review conceptual soundness, data quality, calibration, implementation, sensitivity to assumptions, benchmark models, and performance in stressed periods. Independent validation is essential because a model owner may have incentives to minimise measured capital or reported risk.

6. Stress Testing and Scenario Design

Stress testing explores losses under adverse but plausible conditions. It complements statistical models because historical samples may not contain the relevant combination of shocks. Scenarios may be:

• Historical: replaying events such as a major currency devaluation or rapid rate shock.
• Hypothetical: constructing a coherent scenario, such as a parallel rate increase, currency depreciation, volatility spike, and liquidity widening.
• Reverse stress tests: identifying conditions that would threaten capital, liquidity, or solvency.
• Sensitivity tests: shocking one or a few risk factors without claiming a complete macroeconomic narrative.

A credible scenario specifies risk-factor shocks, horizon, valuation assumptions, management actions, liquidity conditions, and second-round effects. For example, a volatility shock without widening bid–ask spreads may materially understate losses for an options market-maker.

Stress testing should be integrated with capital planning, limit setting, recovery planning, and senior-management reporting. It should not be used merely as a compliance exercise.

7. Hedging and Position Management

7.1 Principles of hedging

A hedge should be evaluated by its economic objective, not by whether one reported sensitivity becomes zero. The manager must identify the exposure, choose an instrument with sufficiently similar risk drivers, specify hedge horizon and rebalancing rules, and monitor residual basis and liquidity risks.

For a linear hedge, the minimum-variance hedge ratio is often written:

where ∆S is the change in the exposure price, ∆F is the change in the hedge-instrument price, Cov denotes covariance, and Var denotes variance. The formula assumes a linear relationship, stable enough historical parameters, and a defined hedge horizon. It minimises variance of the hedged change under the estimated covariance structure; it does not guarantee profit or eliminate tail risk.

7.2 Interest-rate hedging

Interest-rate risk may be hedged with government bonds, futures, interest-rate swaps, swaptions, or combinations of instruments. Duration matching aligns first-order exposure, while key-rate matching addresses curve shape. A duration-neutral portfolio may still have convexity, spread, or basis exposure.

7.3 FX hedging

Forwards and swaps are common tools for managing currency exposure. Options provide asymmetric protection but introduce premium, volatility, and model risk. A hedge policy should distinguish transaction exposure, translation exposure, funding exposure, and trading exposure.

7.4 Equity, commodity, and volatility hedging

Index futures can hedge broad equity beta but leave single-name, sector, dividend, and basis risk. Commodity futures can hedge price risk while leaving location, quality, timing, and roll risk. Volatility books may require dynamic delta hedging plus explicit management of gamma, vega, skew, term structure, and correlation.

Dynamic hedging is vulnerable to transaction costs, discrete rebalancing, jumps, illiquidity, and model error. During a market crisis, the hedge may be most needed precisely when it is most expensive or difficult to execute.

8. Trading-Book Regulation and Governance

The Basel market-risk framework is built around the trading-book boundary, risk-based capital, standardised measurement, internal models subject to supervisory approval, and strong risk governance. Contemporary reforms place greater emphasis on Expected Shortfall for capturing tail risk, liquidity horizons, risk-factor eligibility, desk-level model performance, and the treatment of non-modellable risk factors. Exact implementation varies by jurisdiction and should be read with the relevant supervisor’s rules.

Important governance principles include:

• documented trading intent and boundary controls;
• independent risk control and valuation control;
• prudent independent price verification;
• clear desk-level mandates and limits;
• escalation of limit breaches;
• model approval before production use;
• controls over valuation adjustments, reserves, and uncertainty;
• complete audit trails for data, prices, trades, and overrides.

Capital measures should not be confused with economic risk measures. Regulatory capital is designed partly for comparability and resilience across institutions; an internal risk appetite may be more granular and may include liquidity, concentration, reputational, and strategic considerations.

9. Limitations and Critical Evaluation

No market-risk model observes the future directly. Its weaknesses commonly arise from:

1. Distributional assumptions: returns may be fat-tailed, skewed, or regime-dependent.
2. Correlation instability: diversification can disappear during stress.
3. Liquidity risk: quoted prices may not be executable at the modelled level.
4. Valuation uncertainty: illiquid instruments require judgement and reserves.
5. Data limitations: short histories, stale observations, structural breaks, and survivorship bias distort estimates.
6. Non-linearity: delta-normal approximations can materially understate option losses.
7. Model interaction: pricing, risk, capital, and stress models may share hidden assumptions.
8. Behavioural response: traders may optimise positions against limits rather than economic risk.
9. Procyclicality: risk estimates and capital requirements can rise sharply after volatility increases.
10. Operational complexity: incorrect booking or mapping can produce a precise but meaningless number.

Technology can improve intraday monitoring, scenario generation, data lineage, and anomaly detection. Machine-learning tools may identify nonlinear patterns or unusual trades, but they introduce explainability, stability, bias, and governance concerns. AI should support—not replace—economic interpretation, independent challenge, and accountable decision-making.

10. Illustrative Case: Meridian Bank’s Rates and FX Trading Desk

Background

Meridian Bank operates a rates and FX desk serving corporate and institutional clients. The desk holds government bonds, receives fixed in interest-rate swaps, and sells foreign-currency options. Its daily risk report shows low aggregate delta, but stress losses have increased.

Problem

The apparent neutrality was created by offsetting exposures: long five-year bond duration was offset by short ten-year swap duration; FX spot delta was hedged, but the options book remained short implied volatility. The desk also used a single parallel-shift duration measure and a one-year historical VaR window that contained few crisis observations.

Analysis

Risk management decomposed the portfolio into key-rate duration, swap-spread basis, FX delta, option vega, gamma, and volatility-surface exposures. A 100-basis-point steepening scenario produced a large loss because the five-year and ten-year positions did not offset under a curve twist. A 10-volatility-point increase generated additional losses because the desk was short vega. A widening of bid–ask spreads increased the estimated liquidation loss beyond the market-value stress.

Backtesting showed an acceptable number of ordinary VaR exceptions, but exceptions clustered during volatile weeks. P&L attribution also revealed that the model omitted a material cross-currency basis factor.

Approach and solution

The bank introduced key-rate and basis limits, separate vega and gamma limits, and stress limits based on historical and hypothetical scenarios. It replaced the single VaR view with ES, full-revaluation stress tests, and daily P&L attribution. The desk hedged the curve twist with maturity-specific swaps, purchased limited volatility protection, and reduced positions whose liquidity-adjusted stress losses exceeded appetite.

Outcome

Reported day-to-day risk increased because previously hidden exposures were recognised. However, the portfolio became more resilient: stress losses declined, unexplained P&L fell, and limit breaches were escalated earlier. Management accepted that a higher measured risk number can represent better measurement rather than worse risk.

Learning points

The case demonstrates that aggregate delta can conceal curve, basis, volatility, and liquidity risks. It also shows why VaR exceptions, stress testing, sensitivities, and governance must be interpreted together. Risk management is most effective when measurement leads to specific decisions about positions, hedges, limits, and accountability.

Best-Practice Framework

A mature trading-book framework should:

1. Map every position to economically meaningful risk factors.
2. Use sensitivities for daily steering and full revaluation for nonlinear portfolios.
3. Combine VaR or ES with severe but plausible stress scenarios.
4. Monitor basis, liquidity, concentration, and model uncertainty explicitly.
5. Backtest both exception frequency and exception clustering.
6. Reconcile risk-theoretical, hypothetical, and actual P&L.
7. Set limits at desk, product, factor, tenor, and stress levels.
8. Maintain independent price verification and model validation.
9. Document assumptions, overrides, data sources, and escalation procedures.
10. Review hedges for effectiveness under ordinary and stressed conditions.

Chapter Summary

Market risk in the trading book arises from changes in rates, currencies, equities, commodities, basis relationships, volatilities, and correlations. Sensitivities such as delta, duration, convexity, key-rate duration, gamma, and vega provide transparent local measures, but they can be incomplete for nonlinear or stressed markets. VaR summarises a loss quantile, whereas Expected Shortfall measures the average severity of losses beyond that quantile. Neither replaces stress testing, liquidity analysis, or judgement.

Backtesting evaluates statistical performance, while P&L attribution investigates whether the model captures the drivers of realised outcomes. Hedging requires attention to basis, transaction costs, liquidity, and model risk. Contemporary regulation emphasises trading-book boundaries, tail-sensitive capital measures, liquidity horizons, desk-level controls, and independent governance. The central managerial lesson is that a low headline risk number is not necessarily a safe portfolio: robust management requires decomposition, challenge, scenario analysis, and disciplined action.

Key Terms

Market risk: The possibility of loss from adverse changes in market variables.

Trading book: Positions held for trading or market-making purposes and subject to market-price risk.

Sensitivity: The change in value associated with a specified change in a risk factor.

Duration: A measure of a bond’s first-order sensitivity to yield changes.

Convexity: A measure of the second-order relationship between bond price and yield.

Key-rate duration: Sensitivity to a specific maturity point on the yield curve.

Basis risk: Risk that related prices or rates do not move together as expected.

Delta: First-order sensitivity to an underlying price or risk factor.

Gamma: Sensitivity of delta to the underlying risk factor.

Vega: Sensitivity of an option value to implied volatility.

Value at Risk: A specified loss quantile over a stated horizon and confidence level.

Expected Shortfall: Average loss conditional on losses exceeding a specified threshold.

Backtesting: Comparing model forecasts with subsequent realised outcomes.

P&L attribution: Analysis of the drivers of observed profit and loss.

Stress testing: Evaluation of portfolio performance under adverse scenarios.

Historical simulation: Risk estimation using observed historical market changes.

Liquidity horizon: Time required to exit or hedge a risk position realistically.

Model risk: Risk of loss from incorrect, misused, or poorly implemented models.

Independent price verification: Valuation checking performed separately from the trading function.

Trading-book boundary: The governance and regulatory rules determining whether positions belong to the trading or banking book.

Review Questions

Short Answer Questions

1. Why can a portfolio with zero aggregate duration still have significant interest-rate risk?
2. Distinguish VaR from Expected Shortfall.
3. What does a volatility option’s vega measure?
4. Why is basis risk important in hedging?
5. What is the purpose of P&L attribution?

Long Answer Questions

1. Explain how duration and convexity are used together to estimate bond-price changes. State the assumptions and limitations.
2. Compare historical simulation, parametric VaR, and Monte Carlo VaR.
3. Discuss the role of stress testing alongside VaR or Expected Shortfall.
4. Explain the principal governance requirements of a robust trading-book framework.
5. Critically evaluate the claim that dynamic hedging eliminates option risk.

Analytical / Application Questions

1. A bond portfolio has modified duration 5.0 and convexity 40. Estimate the percentage price change for a 75-basis-point yield increase using the duration–convexity approximation. Explain why a full revaluation may be preferable.
2. A desk is long five-year government bonds and short ten-year interest-rate swaps, with near-zero aggregate duration. Design a diagnostic analysis to determine its exposure to curve shape, spread, basis, liquidity, and stress risk.
3. A one-day 99% VaR model records exceptions that cluster during three volatile weeks. Explain what this indicates, identify further tests, and recommend management actions.

Multiple Choice Questions with Answers

Question 1

Which measure captures the second-order effect of a yield change on a bond’s price?

A. Duration

B. Convexity

C. Vega

D. FX delta

Correct Answer: B. Convexity

Question 2

A 99% one-day VaR of 10 million means that, under the model,

A. the maximum daily loss is 10 million

B. the bank will lose exactly 10 million every 100 days

C. losses greater than 10 million have an estimated probability of about 1%

D. the expected daily loss is 10 million

Correct Answer: C. losses greater than 10 million have an estimated probability of about 1%

Question 3

Expected Shortfall differs from VaR because Expected Shortfall

A. measures the average loss beyond the VaR threshold

B. always assumes normally distributed returns

C. ignores tail losses

D. is only applicable to interest-rate positions

Correct Answer: A. measures the average loss beyond the VaR threshold

Question 4

If modified duration is 4 and the yield rises by 25 basis points, the first-order estimated percentage price change is

A. +1.00%

B. −1.00%

C. +0.25%

D. −0.25%

Correct Answer: B. −1.00%

Question 5

Which exposure is most directly measured by vega?

A. Underlying price

B. Interest-rate level

C. Implied volatility

D. Time to settlement only

Correct Answer: C. Implied volatility

Question 6

A portfolio with offsetting five-year and ten-year exposures may still be vulnerable because

A. aggregate duration always measures curve shape perfectly

B. key-rate exposures can remain non-zero

C. convexity is always zero for bonds

D. basis risk cannot affect rates portfolios

Correct Answer: B. key-rate exposures can remain non-zero

Question 7

The minimum-variance hedge ratio is based primarily on the

A. ratio of means only

B. covariance of exposure and hedge changes relative to hedge variance

C. difference between two VaR figures

D. option premium divided by maturity

Correct Answer: B. covariance of exposure and hedge changes relative to hedge variance

Question 8

Backtesting exception clustering most directly suggests a problem with

A. the independence or conditional calibration of the model

B. the definition of duration

C. the existence of all market risk

D. the currency quotation convention only

Correct Answer: A. the independence or conditional calibration of the model

Question 9

Which is an example of basis risk?

A. A bond’s price falling when its yield rises

B. A hedge instrument failing to move in line with the exposure

C. An option losing time value as maturity approaches

D. A currency being quoted in domestic units

Correct Answer: B. A hedge instrument failing to move in line with the exposure

Question 10

Why is stress testing necessary even when a VaR model backtests successfully?

A. VaR and stress tests always produce identical results

B. Stress tests explore adverse scenarios that may be absent from the historical sample

C. Stress testing removes all model risk

D. VaR measures only accounting profit

Correct Answer: B. Stress tests explore adverse scenarios that may be absent from the historical sample