GARP FRM Part II Market Risk Measurement and Management Study Guide

Quick answer: Study the 18 chapters in four blocks: risk measurement and validation; correlation and dependence; hedging and term-structure models; and volatility surfaces plus FRTB. For every model, learn the required data, calculation sequence, interpretation and failure mode.

The GARP FRM Part II Market Risk Measurement and Management book tests model choice more than formula recall. A candidate must know when historical simulation is suitable, what an exception backtest can and cannot prove, why a Gaussian copula parameter is not observed default correlation, and how an FRTB liquidity horizon changes expected-shortfall aggregation.

This is one printed Part II book. It should be studied as a complete market-risk unit, then combined with the other Part II books and Current Issues for full-exam practice.

Block 1: Risk Measurement and Model Validation

Chapters 1–7 form one continuous workflow:

1. Estimate the loss distribution. Separate P/L from loss/profit data, arithmetic from geometric return, and VaR from expected shortfall.
2. Choose the estimation method. Basic, weighted and filtered historical simulation use empirical data differently. EVT focuses specifically on extreme tails.
3. Backtest forecasts. Exception frequency tests unconditional coverage. Exception clustering tests conditional coverage.
4. Validate the wider model. Conceptual soundness, sensitivity analysis, confidence intervals and benchmarking remain necessary even when exception counts look acceptable.
5. Test the full forecast distribution. Probability integral transforms should be uniform and independent under a correctly specified dynamic distribution.

Keep the limits explicit. Bootstrap resampling does not create events absent from history. A high peaks-over-threshold cutoff leaves fewer observations. Passing a VaR exception test does not validate loss severity beyond the threshold.

Block 2: Correlation, Copulas and Dependence

Chapters 8–10 move from definition to empirical behavior and then model construction.

Correlation measures linear co-movement. It does not establish causation or fully describe dependence. It affects portfolio variance, multi-asset option pricing, correlation swaps, market-risk capital and joint credit losses. During stressed markets, rising correlation can weaken the diversification assumed in ordinary periods.

The empirical chapter adds three practical warnings:

• correlation varies across economic states;
• correlation volatility also changes through time;
• a bounded correlation coefficient should not be fitted carelessly with an unbounded distribution.

Copulas then separate marginal distributions from dependence. In a Gaussian copula, marginal default curves are fitted first, correlated latent variables are generated second, and default times follow by inversion. The latent correlation input is not automatically the observed correlation of default indicators. Copula-family choice matters because joint-tail behavior can change while the marginals remain fixed.

Block 3: Hedging and Term-Structure Models

Chapters 11–16 should be learned as a sequence rather than six isolated readings.

Start with regression hedging. A DV01-neutral position offsets one prescribed parallel-rate sensitivity but can retain slope, curvature and basis risk. Regression beta adjusts the hedge for observed co-movement. A change regression is often more useful for short-horizon hedging than a level regression, and reversing dependent and independent variables generally changes the slope.

Next, use a short-rate tree for arbitrage pricing. Value terminal cash flows, work backward through the tree and distinguish real-world probabilities from risk-neutral probabilities. A replicating portfolio and risk-neutral valuation are alternative routes to the same no-arbitrage price when the model is internally consistent.

Then separate the three forces shaping the term structure:

• expected future short rates;
• compensation for priced interest-rate risk;
• convexity created by rate volatility.

The model chapters build on those forces. Ho–Lee uses time-dependent drift to fit the initial curve. Vasicek adds mean reversion but permits negative rates. CIR uses square-root volatility and is designed for non-negative rates under its parameter restriction. Lognormal models make absolute volatility proportional to the rate level. Gauss+ adds short-, medium- and long-term factors for richer curve behavior.

Block 4: Volatility Surfaces and FRTB

Chapter 17 replaces the single-volatility assumption with observed option-market structure. A smile varies by strike; a skew is asymmetric; a term structure varies by maturity; and a volatility surface combines strike or moneyness with maturity. The surface-movement assumption also affects Greeks. “Hold implied volatility constant” is incomplete unless the candidate states what is held constant—strike, delta or another quoting rule.

Chapter 18 moves to regulatory market-risk capital. Keep the components separate.

Under the standardized approach, delta, vega and curvature sensitivities are combined with default risk and residual risk charges. Under the internal-model approach, expected shortfall replaces VaR, risk factors receive different liquidity horizons, and desk eligibility depends on backtesting and profit-and-loss attribution.

Do not mix legacy VaR rules with FRTB expected shortfall. Do not apply one liquidity horizon to every factor. Do not assume a desk keeps internal-model approval after failed eligibility tests.

A Direct Revision Routine

Use one page per chapter with five lines:

1. Purpose: what the method produces.
2. Inputs: data, parameters, units and horizon.
3. Sequence: the calculation or validation order.
4. Interpretation: what the result means.
5. Failure mode: the assumption most likely to mislead.

Then practise paired distinctions:

• P/L versus loss/profit;
• VaR versus expected shortfall;
• basic versus filtered historical simulation;
• block maxima versus peaks over threshold;
• unconditional versus conditional coverage;
• exception testing versus PIT testing;
• correlation versus independence;
• true versus risk-neutral probability;
• DV01 hedge versus regression/PCA hedge;
• normal versus CIR versus lognormal rates;
• smile versus surface;
• standardized versus internal-model FRTB.

If any pair still sounds interchangeable, return to the chapter before doing more questions.

When to Move to Full Part II Practice

Move on when you can select a method before calculating, explain the limitation after calculating, and connect neighboring chapters without notes. You should be able to diagnose a PIT histogram, explain why a correlation hedge can fail in stress, build a risk-neutral tree, compare short-rate processes and separate the FRTB capital components.

At that point, continue through the remaining Part II books from the GARP FRM exam-preparation hub. Keep Current Issues separate because GARP publishes those required readings online rather than as a sixth printed book.