Study the FRM curriculum by practicing measure selection, not just calculation. For every practice vignette, first name the concept being requested, then compute. Work scenarios where two plausible measures fit the same facts, and record why only one matches the wording. Build a one-page selection card comparing VaR, Expected Shortfall, exposure measures, and stress tests, and test yourself against a rubric until your classification is fast and consistent.
Reading FRM Vignettes: Match the Question to the Named Measure
Each FRM question points to a specific curriculum concept through its wording. Train a two-step habit: identify which named measure or model the stem demands, and only then reach for formulas or numbers.
Curriculum concepts are precise. A question about the average loss given that losses exceed a threshold is asking for Expected Shortfall, not VaR, even though both describe tail risk. A question about potential future counterparty exposure points to PFE or Expected Exposure, not to a current mark-to-market figure. Before calculating, restate the stem in your own words and match it to one concept name from your syllabus list.
Build this habit with your practice bank: for every item, write one line identifying the concept and one line explaining what in the wording signaled it. Signals include phrases like 'average loss beyond', 'loss not expected to exceed at a given confidence level', 'potential exposure at a future date', and 'impact of a severe but plausible move'. Reviewing these signal lines across dozens of items teaches you the vocabulary mapping faster than rereading chapters.
A drill for this section: take ten practice questions and, without solving them, label each with the concept it targets. Check your labels against the answer explanations. If you mislabel more than two, your issue is vocabulary mapping, and more computation practice will not fix it.
- Signal phrases for Expected Shortfall: average, beyond the threshold, conditional on a breach.
- Signal phrases for VaR: minimum loss, at a stated confidence level, over a stated horizon.
- Signal phrases for exposure measures: potential, future date, counterparty, before netting or collateral details.
- Signal phrases for stress testing: severe, plausible, scenario-based, specific market event.
VaR vs Expected Shortfall: Two Different Questions About the Same Tail
VaR reports a minimum threshold loss at a confidence level; Expected Shortfall reports the average loss in the worse-than-VaR region. They answer different questions, so they produce different numbers from the same distribution.
Say a loss distribution puts 5% of outcomes above 100 (in whatever currency units the example uses). The 95% VaR is 100: the smallest loss that the worst 5% of outcomes all meet or exceed. Expected Shortfall at 95% instead asks for the mean of that entire worst-5% region. If losses beyond 100 run to 400, the shortfall will be materially larger than 100, because it averages the severity inside the tail rather than reporting only its edge.
This distinction also explains why the curriculum treats them differently as risk measures. VaR says nothing about how bad losses get once you cross the threshold; two portfolios with identical VaR can have very different tails. Expected Shortfall summarizes the tail's average severity and is described in the curriculum literature as a coherent measure in the standard treatment, meaning it satisfies properties such as subadditivity that VaR can violate for certain portfolios. Compare them on the same sample distribution until both computations feel routine.
A useful contrast line to memorize: VaR is a quantile of the loss distribution; Expected Shortfall is a conditional mean beyond that quantile. Write both definitions, compute both on one five-point loss sample, and note that the shortfall always sits at or above the VaR for the same confidence level in a continuous setting.
Worked Scenario 1: The Tail-Averaging Mistake
A vignette gives five extreme-loss outcomes and asks for Expected Shortfall at 95%. The plausible error is averaging the wrong set of outcomes or including the VaR boundary incorrectly.
Scenario: a portfolio's worst one-day outcomes, ranked worst to best, are 420, 310, 260, 180, and 120, with all other outcomes below 120. The 95% VaR is therefore 120 (the boundary of the worst 5%). Expected Shortfall at 95% is the average of the worst 5% of outcomes: (420 + 310 + 260 + 180 + 120) / 5 = 258. The plausible mistake is reporting 120 because the candidate computed VaR reflexively when the stem said 'average loss in the worst 5%', or averaging only the three outcomes beyond the boundary and ignoring the boundary observation under the discrete-case convention the item's answer choices assume.
The better decision is to match the averaging set to the exact tail the stem defines, and to note which convention the answer choices reflect when the distribution is discrete. This matters because the two numbers, 120 and 258, tell a risk committee very different stories: one says the threshold line, the other says what you should expect to lose on a bad day once that line is crossed. Practicing this on a handful of small samples makes the tail-set choice automatic, and it transfers directly to longer exam-style vignettes that bury the same logic under derivatives positions and market data.
Self-check: on any five-outcome sample, produce VaR and ES in under a minute and state which convention you used for the boundary. If your ES is ever below your VaR at the same confidence level, recheck the averaging set before moving on.
Choosing an Estimation Model: Delta-Normal, Historical Simulation, Monte Carlo
Three estimation approaches dominate the valuation-and-risk part of the curriculum. Each trades off assumptions, data needs, and treatment of nonlinearity, and vignette wording tells you which one fits.
Delta-normal (variance-covariance) methods map positions to linear sensitivities and assume normally distributed factor returns; they are fast but blind to convexity, so they understate risk for portfolios containing options. Historical simulation resamples actual past return vectors, capturing nonlinearity and fat tails in the data without distributional assumptions, but it depends entirely on the chosen lookback window. Monte Carlo simulation generates thousands of scenario paths from specified distributions, handling optionality and complex payoffs flexibly at the cost of model risk and computational burden.
Use the table below as a decision aid. In a vignette, the position description is your cue: a straight bond or simple equity book points toward linear methods; an options-heavy book argues against delta-normal and toward simulation approaches; a question emphasizing 'what actually happened in the recent past' signals historical simulation; a question about pricing path-dependent payoffs signals Monte Carlo. Practicing this cue-to-model mapping is more valuable than memorizing a fourth decimal place of any formula.
When you review answers, write down which sentence in the stem should have triggered your model choice. Over time, the mapping becomes a reflex: 'short-dated options, convex payoff' immediately rules out delta-normal in your head, which is exactly the kind of fast elimination the exam format rewards.
| Approach | Key assumption | Handles optionality? | Main practical limitation |
|---|---|---|---|
| Delta-normal (variance-covariance) | Linear sensitivities; normally distributed factor returns | No, convexity ignored | Understates tail risk for nonlinear portfolios |
| Historical simulation | Past returns represent the near future | Yes, via full repricing or holding-period returns | Blind to events absent from the lookback window |
| Monte Carlo simulation | User-specified distributions and dynamics | Yes, repriced along each path | Results depend on model choices and computational effort |
Worked Scenario 2: Counterparty Exposure and Wrong-Way Risk
Counterparty questions distinguish current exposure, expected exposure, potential future exposure, and wrong-way risk. A vignette about a deteriorating counterparty tests whether you can separate profile shape from probability.
Scenario: a dealer holds an uncollateralized derivative with a counterparty. Current exposure is the positive mark-to-market today, which could be zero even though the contract has years left. Expected Exposure (EE) is the average of possible positive exposures at a future date across simulated scenarios; Potential Future Exposure (PFE) is a high quantile of that same exposure distribution, describing how bad exposure could plausibly get. The plausible mistake in a vignette is quoting today's mark-to-market as the relevant exposure when the stem asks about potential exposure at a future date, ignoring that a zero current value can coexist with a large forward-looking PFE profile that typically rises then falls over the contract's life.
Now add wrong-way risk: exposure to the counterparty increases precisely when the counterparty's default probability increases, for example a firm that has sold protection whose exposure grows as its own credit deteriorates. The better decision is to flag the dependence between exposure and default whenever the vignette links the counterparty's health to the same factors driving the trade, rather than treating exposure and default probability as independent inputs. This distinction matters because mitigation choices differ: collateral and netting address exposure levels, while wrong-way risk may require limits, restructures, or pricing adjustments that generic exposure numbers will not capture.
Practice line: for any counterparty vignette, sketch the EE profile qualitatively (rising then decaying for many products), mark where PFE sits above it, and state whether any fact in the stem creates wrong-way dependence. Three sketches from different product types are enough to make the shape intuitive.
Practice Drill: The Measure-Selection Card and Self-Check Rubric
Build a one-page card mapping stem language to the correct measure, then drill against a rubric. The goal is consistent, fast classification before any arithmetic, checked with concrete observable criteria.
Construct the card in three columns: stem signal, concept, and one-line definition. Include at least VaR, Expected Shortfall, Expected Exposure, Potential Future Exposure, wrong-way risk, stress testing, scenario analysis, and expected versus unexpected loss. Keep definitions to one line each so the card stays scannable. Then drill: take a mixed set of practice items, and for each, cover the answer choices, classify the concept, and only then attempt a solution.
Score each drill item against this rubric, aiming for consistent self-observations rather than a passing prediction: (1) you named the concept before looking at choices, yes or no; (2) your one-line definition matched the answer explanation's logic, yes or no; (3) your computation, when required, used the correct tail set or distributional region, yes or no. Track the rubric across sessions. If columns 1 and 2 are consistently yes but column 3 is not, shift practice time toward computation; if column 1 fails, shift toward vocabulary mapping with more unlabeled classification rounds.
Extend the drill weekly by writing your own two-line vignettes: describe a portfolio and a question in two sentences, then trade mini-vignettes with a study partner or answer them yourself a week later. Writing stems forces you to think from the question-design side, which sharpens your recognition of what each measure can and cannot be asked about.
A Phased FRM Preparation Sequence and Readiness Checks
Structure preparation in four phases: concept vocabulary, computation fluency, integration through mixed sets, and final scenario review. Each phase ends with an observable check before you advance.
Phase one covers the curriculum's topic areas and builds your measure-selection card as you read, so vocabulary mapping develops alongside content. Phase two drills computation on small hand-made samples: VaR and ES on five-outcome lists, exposure sketches, duration-based price change estimates, and simple hedge calculations. Phase three mixes topics in single practice sessions, because integration under mixed conditions is a different skill than blocked topic review. Phase four concentrates on full-length, exam-style sets and your error log, reserving time to reread explanations for every item you classified incorrectly in phase one terms.
Keep an error log with three categories: concept misidentification, computation slip, and reading slip. In the final phase, weight your review by category counts rather than by raw score, because a session that raises your score but leaves concept misidentification untouched has not fixed the costliest pattern. Adapt the phase lengths to your available months and weekly hours; the sequence, not a fixed calendar, is what transfers across candidate situations. If a phase check fails, extend that phase rather than pushing forward, since later phases assume earlier ones hold.
For administrative details such as registration windows, scheduling, fees, and current exam logistics, consult GARP's official pages rather than secondary sources, since those specifics change by sitting.
- Readiness check 1: you can state VaR and Expected Shortfall definitions from memory and compute both on a fresh five-outcome sample without notes.
- Readiness check 2: given any practice vignette, you name the target concept before reading the answer choices, consistently across a mixed set.
- Readiness check 3: you can sketch an Expected Exposure profile and place PFE above it, and state one condition that creates wrong-way risk.
- Readiness check 4: your error log shows concept misidentification shrinking across the final phase, indicating the selection skill is consolidating.
References and further reading
Use these references to explore the concepts and check the latest information from the relevant organizations.