This free guide pairs twenty original multiple-choice questions with sixty technical concepts across the six FRM Part II domains. Practice tail-loss estimation, counterparty exposure, operational resilience, funding stress, portfolio construction and emerging financial risks. GARP describes Part II as eighty equally weighted computer-based questions with four hours allowed; this is a representative preview. For the requested October 16, 2026 scope, supplied evidence identifies the 2026 curriculum but verifies only broad domains, not detailed readings or weights. Citations establish domain relevance, not technical verification. Current Issues examples are illustrative. Required online readings remain examinable; the 2027 registration announcement establishes no replacement syllabus.
What this guide covers
Explore 20 original FRM Part II questions and 60 concepts covering market, credit, liquidity and investment risk.
Market Risk Measurement and Management
1. Loss quantiles and the limits of VaR
Value at Risk is a quantile of a specified loss distribution over a stated horizon. A one-day 99% VaR identifies a cutoff exceeded with approximately 1% probability under the model, rather than a maximum possible loss. Define the loss sign convention and quantile convention before calculating, especially for discrete distributions. VaR summarizes the location of a tail boundary but ignores how severe losses become beyond it. Portfolio VaR also depends on joint behavior, so adding standalone VaRs generally fails to capture diversification or concentration correctly.
Worked example: A desk estimates one-day 99% VaR of USD 4 million. Across many comparable days, its model implies roughly one loss above that amount per hundred observations. A USD 12 million loss is therefore possible without contradicting the definition.
Common mistake: Treating VaR as a worst-case loss turns a probabilistic cutoff into a guarantee that the measure does not provide.
Official syllabus reference: [2]
2. Expected shortfall and fractional tail selection
Expected shortfall averages losses within a specified worst-probability tail. For a continuous distribution, it is commonly expressed as expected loss beyond the corresponding VaR threshold. For discrete distributions, that shortcut can select too much probability; allocate only the necessary fraction of probability mass at the cutoff. Expected shortfall responds to tail severity and satisfies subadditivity under its standard integrable-loss definition. Its estimation remains difficult when tail observations are scarce. Two portfolios can share the same VaR while having very different expected shortfalls because their extreme losses differ.
Worked example: Suppose losses of USD 6 million occupy 3% probability and losses of USD 18 million occupy 1%. The worst 2% includes all severe outcomes and one percentage point of the smaller loss, producing expected shortfall of USD 12 million.
Common mistake: Averaging every observation equal to or above a discrete VaR can include more probability than the intended tail contains.
Official syllabus reference: [2]
3. Threshold exceedances and extreme loss estimation
Peaks-over-threshold modeling separates the probability of crossing a high loss threshold from the distribution of excess losses beyond it. If threshold exceedance probability is p and the desired unconditional tail probability is α, the required conditional excess percentile is 1 − α/p. A fitted generalized Pareto distribution converts that percentile into an excess amount; adding the threshold gives the total loss quantile. For nonzero shape ξ and scale β, the quantile is u + (β/ξ)[(p/α)^ξ − 1]. Threshold choice trades approximation quality against the number of observations available for estimation.
Worked example: Suppose losses exceed USD 8 million with probability 4%, with fitted scale USD 2 million and shape 0.25. A 99% loss quantile requires the 75th excess percentile: 1 − 0.01/0.04. Its value is 8 + 8(4^0.25 − 1) = approximately USD 11.31 million.
Common mistake: Treating the unconditional confidence level as the excess percentile ignores threshold-crossing probability; reporting only the fitted excess also understates the total loss quantile.
Official syllabus reference: [2]
4. Filtered historical simulation
Filtered historical simulation separates changing volatility from standardized return shocks. First estimate historical conditional volatility, then divide each return by its contemporaneous volatility estimate. Rescale these standardized residuals using current or forecast volatility and revalue today's portfolio under the resulting scenarios. This allows historical shock shapes to coexist with current market conditions. The method depends on the quality of the volatility filter and the stability of residual behavior. For multiple risk factors, preserve contemporaneous residual vectors rather than independently shuffling components and accidentally destroying cross-market dependence.
Worked example: A historical return of negative 3% occurred when estimated daily volatility was 1.5%, giving a standardized shock of negative two. With current volatility of 2%, that residual becomes a negative 4% scenario before today's portfolio is revalued.
Common mistake: Rescaling raw returns without first standardizing by their historical volatility can count past volatility differences twice and distort scenarios.
Official syllabus reference: [2]
5. EWMA versus mean-reverting GARCH volatility
EWMA updates variance as a weighted average of yesterday's variance and squared innovation. A larger decay parameter smooths estimates but reacts more slowly to new shocks. Standard GARCH(1,1) adds an intercept and separate lagged variance and squared-innovation coefficients. When their sum is below one under the usual conditions, forecasts revert toward an unconditional variance. EWMA corresponds to a restricted persistence structure without that finite long-run target. Distinguish variance from volatility when calculating updates: average squared quantities first, then take the square root for volatility.
Worked example: Yesterday's variance was 0.0001 and the latest return innovation was 2%. With an EWMA decay parameter of 0.94, updated variance is 0.94 times 0.0001 plus 0.06 times 0.0004, or 0.000118. Updated volatility is approximately 1.09%.
Common mistake: Averaging volatility with a squared return mixes units; the recursion updates variance and requires a final square root.
Official syllabus reference: [2]
6. Coverage and independence in VaR backtesting
Backtesting compares realized losses with forecasts made before those losses occurred. Unconditional coverage asks whether exception frequency matches the nominal tail probability. Independence asks whether exceptions cluster or depend on previous exceptions. Conditional coverage combines these concerns. A model can achieve the correct total count while missing volatility regimes and producing concentrated failures. Specify whether the loss series is actual trading P&L or a hypothetical unchanged-position measure, because intraday activity and fees can affect interpretation. Statistical nonrejection is limited evidence, particularly with short samples and rare exceptions.
Worked example: Two models each produce three exceptions in three hundred observations at a nominal 1% rate. One model's exceptions occur in a single week. The matching counts suggest similar unconditional coverage, but the clustered sequence warrants a separate dependence investigation.
Common mistake: A correct exception count does not establish independence, accurate tail severity, or adequate performance during changing market conditions.
Official syllabus reference: [2]
7. Key-rate sensitivities and yield-curve slope exposure
Key-rate DV01 measures value sensitivity to a localized yield change at a specified maturity. Under the convention that positive DV01 represents a loss when yields rise, approximate value change is minus the sum of each signed DV01 times its yield change in basis points. Opposing maturity exposures can cancel under a parallel shift while remaining sensitive to slope changes. A positive five-year DV01 paired with an equal negative ten-year DV01 gains from five-to-ten-year steepening and loses from flattening. Scenario analysis should therefore retain the maturity pattern rather than relying solely on aggregate DV01.
Worked example: A position has five-year DV01 of +USD 8,000 and ten-year DV01 of −USD 8,000. Five-year yields falling 4 basis points while ten-year yields rise 6 produces −[8,000(−4) − 8,000(6)] = USD 80,000 gain. Reversing those moves causes USD 80,000 loss; parallel sensitivity remains zero.
Common mistake: Zero aggregate DV01 does not eliminate curve risk. Offsetting tenor sensitivities can leave substantial exposure to changes in the yield-curve slope.
Official syllabus reference: [2]
8. Delta-gamma revaluation and nonlinear exposure
A second-order option approximation combines delta times the underlying change with one-half gamma times the squared change. Delta captures local slope; gamma captures changing slope. Positive gamma adds a positive second-order contribution for either direction of underlying movement, holding other inputs fixed. This does not make a long option profitable overall because premium, time decay and volatility changes also matter. A delta-neutral portfolio can retain substantial gamma exposure. Use full revaluation when moves are large, payoffs are discontinuous, or sensitivities change enough to undermine the local approximation.
Worked example: An option has delta 0.40 and gamma 0.06. A USD 2 underlying increase gives a delta contribution of USD 0.80 and a gamma contribution of USD 0.12. The approximate option gain is USD 0.92 before other effects.
Common mistake: Omitting the one-half coefficient doubles the gamma contribution and can materially overstate the benefit of positive convexity.
Official syllabus reference: [2]
9. Volatility smiles, vega and surface movement
Implied volatility is the volatility input that reconciles an option-pricing model with an observed option price. A smile or skew means this input varies across strikes; a term structure means it varies across expiries. Vega measures local price sensitivity to implied volatility under a stated quoting convention. Surface risk requires more than a single parallel volatility shock because skew and maturity segments can move differently. Sticky-strike and sticky-delta assumptions produce different hedging implications when the underlying moves. They are modeling conventions, not guaranteed descriptions of market behavior.
Worked example: A desk owns downside puts and sells near-the-money calls with similar aggregate vega. A steepening downside skew can still increase the puts' implied volatility more than the calls', creating a gain or loss that a parallel-vega report conceals.
Common mistake: Zero net vega under a parallel surface shift does not eliminate exposure to skew changes or different expiry movements.
Official syllabus reference: [2]
10. Serial dependence and horizon scaling
Square-root-of-time volatility scaling follows when returns have constant variance and zero serial covariance over the aggregation horizon. In general, variance of a multi-period sum equals the sum of individual variances plus twice the relevant pairwise covariances. Positive serial dependence increases aggregate variance relative to the independent-return calculation; negative dependence reduces it. Even when volatility scales appropriately, quantile scaling additionally depends on distributional assumptions and portfolio behavior. Option nonlinearities, volatility dynamics and changing positions can invalidate a simple extension of one-day VaR to a longer horizon.
Worked example: Daily return volatility is 1%, and adjacent daily returns have correlation 0.25. Two-day variance is two times daily variance times 1.25. Two-day volatility is therefore approximately 1.58%, rather than the 1.41% obtained by assuming independent returns.
Common mistake: Applying square-root-of-time scaling automatically ignores serial covariance and can understate risk when returns exhibit positive temporal dependence.
Official syllabus reference: [2]
Credit Risk Measurement and Management
11. PD, LGD and exposure at default
Expected credit loss is often approximated as probability of default times loss given default times exposure at default over a consistent horizon. PD measures default likelihood, LGD measures the fraction not recovered, and EAD measures the amount exposed when default occurs. Undrawn commitments may increase EAD through credit conversion factors. The product approximation requires care when exposure, default and recovery depend on common stress conditions; a joint expectation is then more accurate. Expected loss is a distributional mean, whereas unexpected loss concerns variability and tail outcomes around that mean.
Worked example: A USD 6 million drawn loan has a USD 4 million undrawn line with an assumed 50% conversion factor. EAD is USD 8 million. With 2% PD and 40% LGD, the simplified annual expected loss is USD 64,000.
Common mistake: Applying a conversion factor to the already drawn balance understates exposure; it ordinarily applies to the modeled undrawn commitment.
Official syllabus reference: [2]
12. Merton's structural default model
In the basic Merton model, a firm's equity resembles a call option on its asset value with a strike equal to promised debt repayment at maturity. Default occurs at that maturity if assets are insufficient to repay debt. Leverage and asset volatility therefore influence credit risk through the asset-value distribution relative to the debt boundary. Distance to default standardizes the gap between modeled assets and the default threshold. Distinguish physical default probabilities from risk-neutral probabilities used in pricing. The basic model simplifies debt structure and does not naturally capture all early-default mechanisms.
Worked example: Two firms have the same asset value and asset volatility, but one owes substantially more at the modeled debt maturity. Its asset value has a smaller cushion above the repayment boundary, so the model assigns it greater default vulnerability.
Common mistake: Equity volatility is not asset volatility; equity's residual claim and leverage can make its observed volatility substantially higher.
Official syllabus reference: [2]
13. Hazard rates and credit-spread approximations
A hazard rate describes instantaneous default intensity conditional on survival. Under a constant intensity, survival to time t is exp(−λt), so cumulative default probability is one minus that survival probability. A rough credit-spread approximation is spread ≈ intensity times LGD under restrictive pricing assumptions. Market-implied intensity is generally risk-neutral and can incorporate risk premia; observed spreads also reflect liquidity and contractual features. Do not equate a quoted spread directly with default probability. Bootstrapping a term structure from multiple maturities requires consistent discounting, recovery conventions and cash-flow treatment.
Worked example: Assume a constant annual hazard rate of 3%. Two-year survival is exp of negative 0.06, approximately 94.18%, giving cumulative default probability of 5.82%. Simply doubling a one-year probability would neglect the requirement that the borrower first survives.
Common mistake: A credit spread is neither a direct physical default probability nor a pure default signal when liquidity premia are material.
Official syllabus reference: [2]
14. Rating migration and mark-to-market credit loss
A rating transition matrix assigns probabilities of moving from each starting credit state to possible future states, including default. Migration-based portfolio models revalue surviving exposures using the yield or spread associated with the destination state. Loss can therefore occur through downgrade-driven repricing even without default. Under a time-homogeneous Markov assumption, multiplying transition matrices extends the horizon, but changing economic conditions can violate that assumption. Rows must sum to one, and the default state is commonly modeled as absorbing. Recovery valuation applies to defaulted exposures rather than ordinary surviving-state prices.
Worked example: A bond has a 90% probability of retaining its rating, an 8% downgrade probability and a 2% default probability. Its one-year value distribution must include the downgraded bond price, rather than assigning every surviving outcome the unchanged-rating price.
Common mistake: Treating all surviving bonds as unchanged in value misses migration losses and can materially understate mark-to-market credit risk.
Official syllabus reference: [2]
15. Vasicek's common-factor default mechanism
A Gaussian one-factor credit model expresses each borrower's standardized asset return as a common systematic factor plus an idiosyncratic component. Default occurs when this return falls below a threshold calibrated to unconditional PD. Conditional on the common factor, defaults can be modeled as independent, but unconditionally they remain dependent. Greater asset correlation concentrates adverse outcomes when the systematic state is weak. In a highly granular homogeneous portfolio, idiosyncratic variation diversifies and conditional default frequency approaches conditional PD. Finite concentrations, heterogeneous exposures and uncertain recoveries limit that asymptotic simplification.
Worked example: A portfolio contains thousands of similar small loans. A weak common economic factor raises every loan's conditional default probability together. Adding more otherwise identical borrowers reduces borrower-specific noise, but it does not remove the portfolio's exposure to that shared economic state.
Common mistake: Conditional independence given a factor does not mean unconditional default independence; integrating over the shared factor creates dependence.
Official syllabus reference: [2]
16. CreditRisk+ and random default intensity
CreditRisk+ is an actuarial approach that models default counts using Poisson-based machinery, often with random sector intensities to introduce dependence and overdispersion. Exposure amounts are grouped into bands so the aggregate loss distribution can be computed efficiently. Random intensities produce more variable default counts than a fixed-intensity Poisson model and allow common sector conditions to affect many borrowers. The framework emphasizes default losses rather than continuous rating migration. Its approximations require scrutiny for large individual exposures, recovery uncertainty and sectors whose dependence is not represented adequately by the selected factors.
Worked example: A retail portfolio averages ten defaults annually, but defaults surge in recessions. A fixed Poisson count has variance ten. Allowing a shared sector intensity to fluctuate increases count variance and gives more weight to unusually bad default years.
Common mistake: Matching average default frequency with a fixed Poisson distribution does not capture the extra clustering generated by fluctuating economic conditions.
Official syllabus reference: [2]
17. CVA as discounted counterparty loss
Unilateral credit valuation adjustment estimates the discounted loss from a counterparty defaulting while the transaction has positive value to the bank. A simple approximation multiplies LGD by discounted expected positive exposure and marginal default probabilities across future intervals. Expected exposure is a time profile, not simply current mark-to-market. If exposure and default are dependent, use joint modeling rather than multiplying independent averages. Collateral, netting and closeout conventions affect the exposure profile. Bilateral valuation introduces the bank's own default and requires careful treatment of competing default events and valuation conventions.
Worked example: A swap is worth zero today but can become substantially positive next year. Its current exposure is zero, yet its future expected positive exposure can generate positive CVA. Today's fair-value balance alone therefore cannot establish that counterparty credit risk is absent.
Common mistake: Using current exposure at every future date ignores potential future value changes and can severely misstate a derivative's CVA.
Official syllabus reference: [2]
18. Netting sets and collateralized exposure
Closeout netting combines covered transaction values into a single amount following a counterparty default, subject to the enforceability of the agreement. Exposure within an enforceable netting set is based on the positive part of the net value, rather than the sum of each trade's positive value. Collateral can reduce this amount, but thresholds, minimum transfers, disputes and margin delays leave residual exposure. Do not offset across separate counterparties or legally distinct netting sets without an established basis. During the margin period of risk, market movements can create losses before replacement and closeout finish.
Worked example: Two covered trades with one counterparty have values of positive USD 7 million and negative USD 5 million. Enforceable netting reduces current gross positive exposure from USD 7 million to USD 2 million before considering collateral and closeout adjustments.
Common mistake: Offsetting trades solely because they share a corporate group can overstate protection when they belong to different enforceable netting sets.
Official syllabus reference: [2]
19. Wrong-way risk and collateral dependence
Wrong-way risk occurs when counterparty exposure tends to increase as the counterparty becomes more likely to default. The connection can arise from broad market factors or a specific relationship between the trade and counterparty. It can also affect collateral: assets posted as protection may lose value when the pledgor deteriorates. Independent exposure and default models can understate losses under this adverse dependence. Assess the joint stress of transaction value, counterparty condition and collateral realization. Right-way risk describes the opposite tendency, but favorable dependence is not a substitute for sound exposure measurement.
Worked example: A bank receives an airline's own shares as collateral against that airline's derivative obligations. When the airline's finances worsen, its default likelihood rises while the collateral value falls. The collateral therefore offers less protection exactly when protection becomes most important.
Common mistake: A large nominal collateral balance does not establish strong protection if collateral value deteriorates with the counterparty's creditworthiness.
Official syllabus reference: [2]
20. Securitization attachment and detachment points
A securitization tranche absorbs pool losses between its attachment and detachment points. Losses first consume more junior protection; once attachment is breached, additional pool loss reduces tranche principal until detachment is reached. The tranche's percentage loss divides its allocated pool loss by tranche thickness, making thin tranches sensitive to small changes near attachment. Correlation changes the pool-loss distribution and can affect senior and junior tranches differently. Expected pool loss alone cannot determine each tranche's risk. Cash-flow waterfalls, timing, recoveries and structural triggers may complicate the simple loss-allocation calculation.
Worked example: A tranche attaches at 5% and detaches at 10% of a USD 100 million pool. If pool losses reach USD 8 million, the tranche absorbs USD 3 million. Its original USD 5 million principal therefore suffers a 60% loss.
Common mistake: Applying the pool's percentage loss directly to every tranche ignores subordination and the concentration of losses within tranche boundaries.
Official syllabus reference: [2]
Operational Risk and Resilience
21. Inherent and residual risk in RCSA
A risk and control self-assessment maps process risks, evaluates relevant controls and distinguishes inherent exposure from residual exposure after controls. Control design asks whether a control could address the identified failure mechanism; operating effectiveness asks whether it actually functions consistently. A well-designed reconciliation performed irregularly can leave substantial residual risk. Ratings should connect to evidence such as samples, incidents and unresolved exceptions, rather than merely averaging subjective scores. The assessment supports ownership and prioritization, but it does not replace independent challenge or provide a statistically estimated loss distribution by itself.
Worked example: A settlement process has high inherent risk because incorrect instructions can misdirect large payments. Dual approval is suitably designed, but testing finds frequent bypasses. The assessment should therefore recognize ineffective operation rather than assign low residual risk from the written procedure.
Common mistake: A documented control reduces residual risk only to the extent that its design and actual operation address the identified failure mechanism.
Official syllabus reference: [2]
22. KRIs, control indicators and denominator effects
A key risk indicator signals changing exposure or emerging vulnerability, while a control indicator tracks how well a control performs. Useful indicators have a defined population, frequency, owner and escalation response. Rates often interpret changing activity better than raw counts: ten processing errors mean different things across ten thousand and one million transactions. Leading indicators, such as growing unresolved access exceptions, can precede losses; lagging indicators summarize outcomes already realized. Thresholds should reflect the institution's scenario and tolerance, not an assumed universal benchmark, and should be tested for false reassurance and excessive noise.
Worked example: A bank's monthly payment errors fall from twenty to fifteen, but transaction volume falls from two hundred thousand to fifty thousand. The error rate rises from one to three per ten thousand payments, revealing deterioration concealed by the lower count.
Common mistake: A declining incident count can falsely suggest improvement when the underlying activity volume has fallen even faster during the same period.
Official syllabus reference: [2]
23. Compound frequency-severity loss models
A loss distribution approach models aggregate loss as the sum of a random number of event severities. Under independence between frequency and identically distributed severities, expected aggregate loss equals expected frequency times expected severity. For Poisson frequency, aggregate variance equals expected frequency times the second moment of severity. Heavy-tailed severities can make extreme quantiles highly sensitive to limited observations and distribution choice. Reporting thresholds, inflation, changing business scale and missing small losses affect calibration. Dependence across event types or common shocks must be considered when aggregating business units into an enterprise distribution.
Worked example: Assume three incidents per year on average and mean incident severity of USD 80,000. Expected annual loss is USD 240,000. This mean does not determine a high-percentile loss because different severity tails can produce the same average.
Common mistake: Multiplying an annual frequency quantile by a severity quantile generally does not produce the corresponding aggregate annual loss quantile.
Official syllabus reference: [2]
24. Scenario analysis for rare operational events
Operational scenario analysis examines plausible severe failures that internal history may not contain. A useful scenario specifies an initiating event, failed or stressed controls, affected processes, duration and loss channels. Experts should distinguish direct costs, remediation, interruption and potential recoveries without double counting. Elicited likelihoods and severities require challenge because availability, anchoring and optimism biases affect judgments. Scenarios complement internal and external data rather than automatically replacing them. They can expose control dependencies and support stress decisions even when probability estimates remain too uncertain for precise statistical calibration.
Worked example: A workshop models a corrupted customer database that propagates into backup copies. The team estimates restoration effort, customer remediation and service interruption separately, then checks whether the same staff and outage costs have been counted under more than one category.
Common mistake: A severe scenario is not automatically a forecast; assigning precise probabilities without defensible evidence can conceal the uncertainty it was meant to explore.
Official syllabus reference: [2]
25. Insurance deductibles, limits and retained loss
Operational risk insurance transfers specified losses subject to contractual coverage. For a simple policy, payment is the smaller of the per-event limit and the positive amount by which covered loss exceeds the deductible. Retained loss equals gross loss minus actual recovery. Annual aggregate limits, exclusions, waiting periods and insurer default can further constrain protection. Insurance changes the retained severity distribution, but it does not inherently change incident frequency or restore interrupted services. Evaluate recovery timing as well as amount because a valid claim may still leave an immediate liquidity need.
Worked example: A covered USD 900,000 event has a USD 100,000 deductible and a USD 500,000 maximum insurer payment. Although loss above the deductible is USD 800,000, recovery is capped at USD 500,000, leaving the firm to retain USD 400,000.
Common mistake: Subtracting only the deductible from every event overstates insurance protection whenever policy limits, exclusions or aggregate exhaustion restrict recovery.
Official syllabus reference: [2]
26. Accountability in the three lines model
The three lines model distinguishes business ownership of risk, management oversight and independent assurance. Operational teams own the processes and controls they execute. Risk and compliance functions provide expertise, frameworks, monitoring and challenge, while internal audit independently assesses governance and effectiveness. The governing body oversees accountability and receives assurance. Effective separation concerns responsibilities and independence rather than a rigid organizational chart. An oversight function that operates the control it later evaluates faces a self-review problem. Escalation should identify a decision owner rather than diffuse responsibility across committees and assurance teams.
Worked example: A lending team checks borrower documentation, a credit risk function challenges the policy and exception trends, and internal audit tests whether the framework works. If checks fail, the lending team repairs its process while oversight and assurance retain their separate roles.
Common mistake: Independent challenge does not transfer ownership of a business control to the risk function or make internal audit responsible for daily execution.
Official syllabus reference: [2]
27. Service resilience and impact tolerances
Operational resilience focuses on sustaining important services through disruption. Map each service to the people, technology, facilities, information and third parties required to deliver it. An impact tolerance defines the disruption the organization is prepared to withstand for that service, using relevant dimensions such as duration, affected users or transaction backlog. A technology recovery objective addresses a supporting component and may not cover end-to-end restoration. Test severe but plausible scenarios across dependencies, including reconciliation and backlog processing. Successful component restart does not establish that customers can use the service again.
Worked example: A securities platform restarts within one hour, but stale positions require three more hours of reconciliation before trading resumes. A two-hour service disruption tolerance is breached even though the infrastructure team achieves its one-hour server recovery objective.
Common mistake: Equating server availability with service recovery overlooks the operational steps and external dependencies required to restore usable customer outcomes.
Official syllabus reference: [2]
28. Cyber segmentation and recoverable backups
Cyber controls address different stages of an attack. Strong authentication limits unauthorized access, least privilege restricts what an account can do, and segmentation reduces lateral movement across systems. Monitoring supports detection, while isolated or immutable backups support recovery from corruption or ransomware. Backup existence alone is insufficient: restoration must be tested against realistic data volumes and clean-system requirements. Recovery point objectives concern acceptable data loss, while recovery time objectives concern restoration speed. A robust design combines these mechanisms because prevention can fail and a fast restart can still restore compromised or incomplete data.
Worked example: An attacker compromises one employee account but encounters restricted permissions and a separate production network. If destructive activity still reaches a database, an independently protected backup offers recovery. A successful restoration test must also verify data integrity and transaction completeness.
Common mistake: Backups stored with the same credentials and writable network access as production can be compromised by the same incident.
Official syllabus reference: [2]
29. Third-party concentration and substitutability
Third-party risk extends beyond an individual supplier's service quality to shared dependencies and the feasibility of replacement. Several vendors may rely on the same cloud provider, telecommunications route or specialist subcontractor, creating hidden concentration. Assess critical services, data access, incident notification, recovery arrangements and the practical time needed to exit. A contractual termination right does not create operational substitutability if data formats, staff expertise or infrastructure prevent migration. Test exit and fallback arrangements at meaningful scale, and evaluate correlated disruption rather than assuming supplier diversification automatically provides independent resilience.
Worked example: A bank contracts with three payment applications and considers itself diversified. Dependency mapping shows that all three use the same regional cloud service. A regional outage could therefore disable every application despite separate contracts and different vendor brands.
Common mistake: Counting direct suppliers overstates diversification when those suppliers share a critical underlying infrastructure provider or cannot be replaced quickly.
Official syllabus reference: [2]
30. Model validation and effective challenge
Model risk arises when decisions rely on incorrect models, inappropriate assumptions, implementation errors or use outside intended boundaries. Validation should assess conceptual soundness, data quality, implementation and outcomes, with independence sufficient for effective challenge. Benchmarking asks how results compare with alternatives; backtesting compares prior predictions with subsequent observations. Neither alone establishes suitability for every use. Maintain an inventory linking models to owners, limitations and material decisions. Changes to data, methodology or deployment conditions can require renewed assessment. A well-validated model can still be misused through unsupported extrapolation or ignored limitations.
Worked example: A pricing model reproduces benchmark prices for ordinary bonds but is deployed on instruments with embedded options it cannot represent. Implementation accuracy has not solved conceptual mismatch. Validation should identify the unsupported use and require an appropriate valuation approach.
Common mistake: A model that calculates its documented equations correctly can still be unsuitable if those equations omit features material to the decision.
Official syllabus reference: [2]
Liquidity and Treasury Risk Measurement and Management
31. Liquidity coverage and capped cash inflows
A liquidity coverage measure compares eligible high-quality liquid assets with modeled short-horizon net cash outflows. Where the applicable framework caps recognized inflows, the denominator is outflows minus the lesser of eligible inflows and the permitted cap. This prevents a bank from relying entirely on expected receipts while holding little liquid stock. Eligibility, valuation haircuts, operational availability and restrictions determine which assets count. Use the supplied framework and assumptions rather than inventing universal local requirements. A reported ratio also does not establish that cash is available in the correct currency, entity or time window.
Worked example: In an exercise, outflows are USD 100 million and inflows USD 90 million, with recognized inflows capped at 75% of outflows. Net outflows are USD 25 million. Eligible liquid assets of USD 40 million therefore produce a 160% ratio.
Common mistake: Subtracting all projected inflows ignores an applicable inflow cap and can make short-horizon liquidity coverage appear substantially stronger.
Official syllabus reference: [2]
32. Stable funding and balance-sheet structure
A stable funding measure compares available stable funding with required stable funding across the balance sheet and selected commitments. Funding receives weights reflecting stability and maturity, while assets receive required-funding weights reflecting liquidity and tenor under the applicable framework. This structural perspective differs from a short-horizon cash stress ratio. A long-lived illiquid asset generally demands more reliable financing than an immediately monetizable asset. Ratios depend on classification and behavioral assumptions, so examine economic substance and concentration. Improving a weighted ratio need not resolve intraday cash shortages or entity-specific transfer restrictions.
Worked example: For a hypothetical exercise, stable funding is USD 90 million and required stable funding USD 100 million. Replacing an asset that requires USD 17 million of funding with one requiring USD 10 million lowers total requirements to USD 93 million.
Common mistake: A stronger structural funding ratio does not guarantee sufficient immediately available cash during a concentrated short-term withdrawal event.
Official syllabus reference: [2]
33. Cumulative cash-flow ladders and survival horizon
A cash-flow ladder places contractual and behavioral receipts and payments into time buckets, then tracks cumulative available liquidity. The survival horizon is the point at which the modeled liquidity buffer is exhausted under specified stress assumptions. Include contingent draws, margin calls, restricted balances and realistic asset monetization timing. A positive final balance can conceal an earlier cash deficit, so inspect every relevant bucket. Avoid counting an asset both as a sale receipt and as remaining collateral for new borrowing. Currency and legal-entity constraints can prevent apparently offsetting cash flows from meeting obligations.
Worked example: A firm starts with USD 30 million, pays USD 20 million on Monday and USD 25 million on Tuesday, then receives USD 30 million Wednesday. Its Wednesday balance is positive, but Tuesday produces a USD 15 million funding gap.
Common mistake: Netting a week's total receipts and payments can hide an earlier default-causing shortage before later cash receipts become available.
Official syllabus reference: [2]
34. Repo haircuts and deleveraging feedback
A repo haircut limits the loan to a fraction of collateral market value. Falling prices and increasing haircuts both reduce borrowing capacity, creating a need for cash or additional collateral. Selling pledged assets to repay borrowing also removes collateral, so the sale amount needed can exceed the initial margin gap. Forced sales may depress prices further, producing a funding-market feedback loop. Distinguish market-value losses from haircut changes and calculate both. The available response depends on sale liquidity, collateral eligibility, substitution rights and whether financing can be renewed rather than merely on accounting net worth.
Worked example: Collateral worth USD 50 million supports a USD 45 million loan at a 10% haircut. If its value falls to USD 45 million and the haircut becomes 20%, permitted borrowing falls to USD 36 million, creating a USD 9 million repayment need.
Common mistake: Assuming the old haircut still applies after a price shock overlooks lenders' ability to reduce advance rates during stress.
Official syllabus reference: [2]
35. Matched-maturity funds transfer pricing
Funds transfer pricing allocates funding costs and benefits internally so business decisions reflect liquidity consumption and provision. Matched-maturity pricing charges a long-term asset for funding with comparable tenor rather than today's cheapest overnight rate. Stable deposits can receive an internal funding benefit, adjusted for their expected behavior and servicing costs. Contingent commitments may need liquidity charges even before they are drawn. Separate the customer's credit spread from the institution's funding and liquidity costs where possible. The methodology should avoid rewarding business units for maturity transformation whose risk is borne centrally by treasury.
Worked example: A desk originates a five-year fixed-rate loan and proposes charging it the overnight funding rate. Treasury instead assigns a five-year internal funding curve and a relevant liquidity charge, allowing the desk's apparent margin to reflect the financing commitment it creates.
Common mistake: Using short-term marginal funding costs for every asset can make long-term lending appear profitable while concealing its refinancing exposure.
Official syllabus reference: [2]
36. Deposit beta, decay and behavioral repricing
Deposit beta measures how much a deposit rate changes relative to a reference market rate. A 40% beta implies a 40-basis-point deposit repricing for a 100-basis-point reference move, under the modeled conditions. Deposit decay describes balance runoff, while migration describes shifts among account types or competing products. These mechanisms jointly affect funding cost and stability. Historical betas can change with competition, customer segmentation and rate levels. Nonmaturity deposits therefore need behavioral assumptions for earnings and economic-value analysis rather than being treated as either permanently free funding or immediately repricing wholesale liabilities.
Worked example: A bank has USD 200 million of deposits. A one-percentage-point market rate increase and a 25% assumed beta raise annual interest expense by USD 500,000 if balances remain fixed. Customer migration could raise the actual cost beyond that simple estimate.
Common mistake: A low observed deposit beta does not imply permanent funding stability when customers can withdraw or migrate to higher-paying products.
Official syllabus reference: [2]
37. Economic value versus net interest income
Net interest income sensitivity examines how rate changes affect interest earnings over a chosen accounting horizon. Economic value sensitivity examines the present value of future asset, liability and off-balance-sheet cash flows. The measures can move differently because near-term repricing and long-term duration exposures are not the same. Behavioral deposit assumptions, prepayments and embedded options influence both. A duration approximation can estimate small parallel-shock economic-value changes, but nonlinear instruments and nonparallel curves require richer analysis. An institution should understand which risk horizon a proposed hedge improves and whether it worsens the other measure.
Worked example: A bank funds long-term fixed-rate loans with deposits that reprice slowly. A rate increase may initially improve earnings if some assets reset faster, yet reduce economic value when the long-duration asset portfolio loses more value than the modeled liabilities.
Common mistake: A favorable one-year earnings sensitivity does not establish favorable economic-value sensitivity across the full remaining life of the balance sheet.
Official syllabus reference: [2]
38. FX swaps and currency-specific funding
An FX swap exchanges currencies now and reverses the exchange at an agreed future rate, allowing one currency's funding to be transformed into another for a stated tenor. It can hedge exchange-rate exposure while leaving rollover and collateral liquidity needs. Cross-currency basis reflects deviations from a simple frictionless funding relationship and can widen when demand for a currency intensifies. Aggregate cash in the reporting currency can hide shortages in another currency or legal entity. Stress testing should include swap maturity, market access, margin requirements and the availability of usable funding in the required currency.
Worked example: A firm borrows euros and obtains dollars through a three-month FX swap to finance a one-year dollar asset. The exchange rate for the initial swap is fixed, but dollar financing beyond three months still depends on renewing or replacing the swap.
Common mistake: Hedging exchange-rate movements with a short FX swap does not eliminate the refinancing risk of a longer-lived foreign-currency asset.
Official syllabus reference: [2]
39. Intraday liquidity and payment sequencing
Intraday liquidity concerns the ability to meet obligations at the time they fall due within a day. End-of-day net cash can be adequate while morning payment obligations exceed available balances. Payment sequencing, settlement-system rules, collateralized credit and incoming-payment uncertainty determine peak funding needs. Delaying one institution's payments can also delay counterparties' receipts and create gridlock. Measure gross flows and their timing, identify time-critical obligations, and distinguish reusable intraday facilities from committed overnight funding. A daily liquidity report that only records closing balances cannot reveal the largest within-day cash requirement.
Worked example: A bank must pay USD 80 million at ten in the morning but expects USD 90 million at three in the afternoon. With only USD 20 million opening cash, it needs USD 60 million of timely intraday liquidity despite a positive closing balance.
Common mistake: A positive closing cash balance does not demonstrate that every earlier payment could have been settled when required during the day.
Official syllabus reference: [2]
40. Contingency funding actions and execution capacity
A contingency funding plan connects stress indicators with executable liquidity actions, decision authority and communication responsibilities. Potential actions include monetizing eligible securities, drawing reliable facilities, changing funding tenor and reducing cash-intensive activity. Their value depends on timing, legal availability, collateral preparation and market conditions. A list of possible sources is insufficient if all rely on the same stressed market or the same already-pledged assets. Operational tests should confirm access and settlement capability. Separate institution-specific stress from market-wide stress because an action feasible during a local problem may fail when many firms seek funding simultaneously.
Worked example: Treasury lists central-bank-eligible securities as a contingency source, but a test reveals that the assets sit in an entity without completed access arrangements. The plan must reflect the transfer and operational lead time before treating that funding as immediately available.
Common mistake: Counting untested facilities and already-encumbered collateral as immediately usable liquidity exaggerates the resources available during the first hours of stress.
Official syllabus reference: [2]
Risk Management and Investment Management
41. Mean-variance optimization and estimation error
Mean-variance optimization selects portfolio weights using expected returns, covariances and constraints to trade expected reward against variance. Diversification arises from imperfect covariance, not simply from holding many securities. Expected returns are especially difficult to estimate, and small input changes can produce extreme or unstable optimal weights. Constraints, covariance shrinkage and robust approaches can improve practical stability, though they introduce their own assumptions. Variance treats upside and downside deviations symmetrically and does not fully describe skewed or illiquid returns. Evaluate the economic plausibility and sensitivity of a solution rather than accepting mathematical optimality as evidence of reliable performance.
Worked example: An optimizer assigns almost the entire portfolio to one asset after its estimated expected return rises slightly. The analyst checks input uncertainty and adds realistic concentration constraints, finding that the original apparent advantage was too fragile to justify the allocation.
Common mistake: An optimizer's precise weights do not imply precise knowledge of future returns; unstable estimates can generate confidently misleading allocations.
Official syllabus reference: [2]
42. CAPM beta and Jensen's alpha
CAPM beta measures covariance with the market divided by market variance. Under the model, expected excess return equals beta times the market risk premium. Jensen's alpha compares realized or estimated portfolio return with that beta-adjusted benchmark. A positive alpha estimate can reflect skill, sampling noise, omitted factors or an unsuitable market proxy. Beta describes systematic exposure rather than total volatility, so low-beta assets can still have substantial idiosyncratic risk. Use consistent horizons and risk-free rates, and distinguish ex post performance measurement from a justified forecast of persistent manager ability.
Worked example: Assume a 2% risk-free rate, an 8% market return and portfolio beta of 1.2. The CAPM benchmark is 9.2%. A portfolio return of 10% produces an estimated alpha of 0.8 percentage points over the measured period.
Common mistake: Subtracting the market return directly from portfolio return is not Jensen's alpha unless the relevant beta adjustment happens to make it equivalent.
Official syllabus reference: [2]
43. Information ratio and active risk
The information ratio divides benchmark-relative return by tracking error, the volatility of active returns. It evaluates the consistency of return relative to a chosen benchmark rather than excess return over cash. Benchmark selection is therefore central: an inappropriate benchmark can reward structural exposures as apparent skill. Match the return and risk horizons and understand whether active return is arithmetic or modeled expected return. A high historical information ratio can be unstable in a short sample. The fundamental law links expected information ratio to skill and effective breadth under restrictive assumptions about independent decisions and implementation.
Worked example: A manager earns 9% while the benchmark earns 6%, with annualized tracking error of 5%. The information ratio is 0.60. If the risk-free rate is 2%, that value still cannot be interpreted as a Sharpe ratio without total volatility.
Common mistake: Using total portfolio volatility in the denominator changes the question from benchmark-relative efficiency and does not produce the information ratio.
Official syllabus reference: [2]
44. Euler allocation of portfolio volatility
Euler allocation attributes portfolio volatility to holdings through their marginal effects on total risk. With weights w, covariance matrix Σ, and portfolio volatility σp = √(wᵀΣw), asset i contributes RCi = wi(Σw)i/σp. These contributions sum to portfolio volatility, while dividing each contribution by σp gives its risk share. Capital weights alone cannot establish those shares: individual volatility and covariance with the portfolio also matter. A small allocation to a volatile asset can dominate total risk. Negative contributions are possible when a holding offsets other exposures, making the decomposition useful when evaluating hedges and risk budgets.
Worked example: With weights 80%/20%, volatilities 5%/40%, and zero correlation, variance is 0.8²×0.05² + 0.2²×0.40² = 0.008; volatility is 8.944%. Euler contributions are 0.0016/0.08944 ≈ 1.789 percentage points and 0.0064/0.08944 ≈ 7.155 points. The 20% holding supplies 80% of total volatility.
Common mistake: An asset's standalone volatility times its weight is generally not its Euler contribution; covariance with the complete portfolio determines its marginal effect.
Official syllabus reference: [2]
45. Factor models and hidden common exposures
A factor model separates returns into exposures to common drivers and a residual component. Portfolio factor exposure is the weighted sum of asset or manager exposures, while factor covariance determines how those common exposures interact. This reveals concentrations concealed by asset labels or manager counts. Return attribution asks which factors contributed to performance; risk attribution asks how factor and residual variation contribute to uncertainty. Estimated factors and loadings can change across regimes. Residuals are not automatically independent, and an omitted common driver can make apparently diversified strategies fail together.
Worked example: An investor hires separate equity, credit and alternative managers. A factor analysis finds that all perform poorly when funding spreads widen. The investor has diversified manager names while retaining a substantial common liquidity-sensitive exposure across the combined portfolio.
Common mistake: Different strategy labels do not establish independent economic exposures; shared factors can dominate the combined portfolio despite organizational diversification.
Official syllabus reference: [2]
46. Futures overlays for beta targeting
An equity futures overlay changes systematic market exposure without immediately trading the underlying holdings. The approximate contract count equals portfolio value times the difference between current and target beta, divided by futures contract notional and the contract's relevant beta. Selling reduces positive market exposure; buying increases it. The hedge depends on stable estimated relationships and can leave basis risk if the portfolio differs from the index. Variation margin creates liquidity requirements even when the economic hedge works. Rebalance as portfolio value, contract value, beta estimates and investment objectives change.
Worked example: A USD 50 million portfolio has beta 1.1 and targets beta 0.5. With USD 200,000 index futures notionals and unit index beta, selling 150 contracts removes USD 30 million of index-equivalent exposure while preserving the original stock holdings.
Common mistake: Hedging the full current beta when the target remains positive removes too much market exposure and changes the intended allocation.
Official syllabus reference: [2]
47. Redemption terms and asset liquidity mismatch
An investment vehicle faces liquidity mismatch when investors can redeem faster than underlying assets can be sold at reasonable cost. Meeting early withdrawals through liquid holdings can leave remaining investors with a more concentrated illiquid portfolio. Selling illiquid assets under pressure can impose market impact and transfer costs across investors. Analyze stressed redemption demand, sale capacity and settlement timing together. Contractual tools such as notice periods or redemption limits must be evaluated within the vehicle's actual terms and applicable framework. Accounting valuation stability does not establish the ability to turn assets into cash promptly.
Worked example: A fund offers daily withdrawals but owns loans that usually take weeks to sell. Initial redemptions consume its government bills. Remaining investors then hold a portfolio with a higher loan concentration, making the next withdrawal wave harder to meet.
Common mistake: A low-volatility reported NAV does not prove liquidity when underlying positions are infrequently priced or costly to sell.
Official syllabus reference: [2]
48. CPPI cushions and gap risk
Constant proportion portfolio insurance allocates to risky assets as a multiple of the cushion, defined as portfolio value minus a designated floor. As losses shrink the cushion, the strategy reduces risky exposure; gains allow greater exposure. The floor is a strategy objective rather than an unconditional guarantee. A sufficiently large price jump between rebalancing opportunities can exhaust the cushion before the portfolio trades. Leverage constraints, transaction costs and market closures also affect implementation. The multiplier determines sensitivity to market moves and the severity of potential gap risk, not merely expected return.
Worked example: A portfolio worth USD 100 has a floor of USD 90 and a multiplier of four, implying USD 40 of risky exposure. A sudden 30% risky-asset loss before rebalancing removes USD 12, taking total wealth below the USD 90 floor.
Common mistake: The CPPI floor can be breached by discontinuous market moves; continuous rebalancing assumptions do not make real-world capital protection certain.
Official syllabus reference: [2]
49. Drawdowns and path-dependent investment risk
A drawdown measures the decline from a prior portfolio peak to a subsequent value. Maximum drawdown is the largest such peak-to-trough decline over the measurement window. Unlike variance, it depends on return ordering and the chosen sample period. Recovery requires a larger percentage gain than the preceding percentage loss because gains apply to a smaller base. Drawdown limits can influence investor behavior and funding decisions, but a historical maximum is not a bound on future losses. Compare strategies using consistent observation frequency, valuation quality and treatment of external cash flows.
Worked example: An investment falls from USD 100 to USD 75, a 25% drawdown. Returning to USD 100 requires a gain of USD 25 on USD 75, or 33.3%. Two portfolios with identical average returns can have different drawdowns because their return sequences differ.
Common mistake: A percentage gain equal to a preceding percentage loss does not restore wealth because the gain starts from a reduced capital base.
Official syllabus reference: [2]
50. Survivorship, backfill and smoothed returns
Investment databases can overstate performance when failed funds disappear, creating survivorship bias. Backfill bias arises when managers add favorable historical records after choosing to enter a database. Infrequent or discretionary valuation can smooth reported returns, understating volatility and creating serial correlation. These distortions affect performance ratios, diversification estimates and allocation decisions. Use dead-fund data where available, distinguish genuinely contemporaneous observations from backfilled history, and inspect valuation practices. Smoothing adjustments require assumptions and cannot fully reconstruct unobserved market prices. A clean-looking track record can therefore reflect reporting mechanics as well as investment outcomes.
Worked example: A database shows an attractive average return for funds operating today. Including funds that closed after large losses reduces that average. Separately, a private-asset fund's quarterly appraisals may make its reported volatility look lower than the economic risk of its holdings.
Common mistake: Low reported volatility is not conclusive evidence of low economic risk when stale prices and selective database inclusion affect observations.
Official syllabus reference: [2]
Current Issues in Financial Markets
51. Physical climate hazards and financial transmission
Physical climate risk transmits through hazards, exposure and vulnerability. Acute hazards include discrete damaging events, while chronic changes can alter productivity, operating costs and asset usability over time. A hazard becomes financially relevant through the location and resilience of people, assets and supply chains. The resulting losses can appear as credit deterioration, insurance claims, collateral impairment or business interruption. Geographic averages may hide highly concentrated local exposure. Distinguish a modeled scenario from a probability forecast, and avoid assuming that insurance fully offsets losses when coverage, renewability and recovery timing may change.
Worked example: Two warehouses face the same modeled flood depth, but one has elevated electrical systems and protected access routes. Their physical hazard is similar, yet operational interruption and repair costs can differ materially because vulnerability and adaptation are different.
Common mistake: Hazard intensity alone does not determine financial loss; exposure, vulnerability, adaptation and contract terms affect the resulting economic damage.
Official syllabus reference: [1]
52. Transition risk and stranded-asset economics
Transition risk arises as policy, technology, market preferences and production systems change during decarbonization. An asset can become economically stranded when expected future cash flows no longer justify its carrying or investment value. Analyze both direct costs, such as emissions charges, and indirect effects through demand, input prices and competition. High current emissions do not mechanically imply the highest credit loss: pricing power, adaptation investment and contractual pass-through matter. Financial transmission can include earnings pressure, refinancing difficulty and collateral repricing. Scenario assumptions should specify the timing and speed of adjustment rather than treating transition as a single uniform shock.
Worked example: A manufacturer faces a new emissions cost but can pass most of it to customers under existing contracts. A competitor with fixed selling prices cannot. Similar emissions intensity therefore produces different margin compression and different implications for debt-service capacity.
Common mistake: Ranking borrowers only by emissions ignores their pricing power, adaptation options and financing structure, which determine how transition shocks affect credit.
Official syllabus reference: [1]
53. Climate scenarios, horizons and balance-sheet assumptions
Climate scenario analysis traces a specified economic and physical pathway into financial outcomes. Long horizons reveal structural adjustment but increase uncertainty about technology, policy and business strategy. A static balance sheet holds exposures broadly fixed to isolate vulnerability; a dynamic balance sheet permits modeled adaptation, lending changes or portfolio turnover. The latter can appear safer simply because optimistic management actions are assumed. Separate scenario severity from scenario probability, and distinguish exploratory analysis from short-horizon forecasting. Transparent transmission assumptions matter more than presenting distant loss estimates with unwarranted numerical precision.
Worked example: A bank projects losses on a twenty-year transition pathway. Under a static balance sheet, carbon-intensive exposures remain. Under a dynamic version, the bank assumes borrowers refinance and its own lending mix changes. Comparing the results requires explicit disclosure of those assumed actions.
Common mistake: A lower dynamic-scenario loss does not demonstrate lower current vulnerability when the reduction depends on untested future management actions.
Official syllabus reference: [1]
54. Temporal leakage and overlapping financial labels
Temporal leakage occurs when model development uses information unavailable at the intended prediction time. It can enter through revised economic data, publication delays, preprocessing fitted on the full sample or labels spanning future intervals. With overlapping forward-return or credit-event windows, random splits can place related outcomes in training and testing. Chronological evaluation and removal of training observations whose label windows overlap the test period reduce this contamination. Feature construction must also respect real availability timestamps. Strong validation performance is informative only when the evaluation reproduces the information constraints of actual deployment.
Worked example: A model uses daily observations to predict the next month's credit deterioration. Training labels from late March include April outcomes, while April observations form the test set. Removing those overlapping training labels helps prevent the model from indirectly seeing test-period information.
Common mistake: Chronological splitting alone is insufficient when training labels extend into the test period or data were revised after the prediction date.
Official syllabus reference: [1]
55. AI explanations, correlated features and causality
Model explanations describe how a fitted system associates inputs with predictions; they do not automatically identify causal effects. Global importance summarizes behavior across observations, while local explanations address a particular prediction. Correlated features can share or redistribute apparent importance depending on the explanation method and background dataset. An explanation can look plausible while the model relies on unstable proxies. Evaluate explanation stability, sensitivity and consistency with the intended use. Decisions about interventions require causal reasoning beyond predictive attribution, especially when changing one input would also change related variables or the population being modeled.
Worked example: An explanation assigns strong importance to a customer's account balance and little to income, although the two are highly correlated. Removing balance may shift importance to income. The original ranking does not prove that changing balance itself causes lower default risk.
Common mistake: Feature importance measures predictive attribution under a method's assumptions; they do not establish the effect of intervening on that feature.
Official syllabus reference: [1]
56. Generative AI, retrieval and unsupported outputs
Generative AI produces plausible sequences rather than guaranteeing factual correctness. Retrieval can supply relevant documents, but it does not ensure the final response faithfully represents them. A financial deployment should separate retrieved evidence from operating instructions, preserve source provenance and test whether material claims are actually supported. Prompt injection arises when untrusted content attempts to redirect system behavior. Permission boundaries, restricted tool access and output validation reduce the impact of such manipulation. Evaluate task-specific error severity, not merely fluent style or average benchmark scores, particularly when generated text feeds subsequent financial decisions.
Worked example: A risk assistant retrieves a borrower report containing an instruction to omit adverse findings. The report should supply evidence about the borrower, while the application's trusted rules continue to determine how evidence is assessed and what actions the assistant can take.
Common mistake: Retrieval reduces some knowledge gaps but does not guarantee faithful reasoning, correct citations or resistance to instructions embedded in documents.
Official syllabus reference: [1]
57. Stablecoin reserves and par-redemption pressure
A reserve-backed stablecoin's stability depends on reserve value, liquidity, custody arrangements and the actual redemption claim. Assets recorded near par can still fall in market value or take time to sell. If liabilities remain redeemable at par, early redemptions can consume liquid reserves and leave remaining holders with weaker coverage. Distinguish market-value solvency from immediate cash availability: an issuer can face either problem or both. Redemption rights and legal protections vary by arrangement, so no universal guarantee should be assumed. Stress reserve prices, settlement timing and concentrated withdrawal demand together.
Worked example: An issuer has USD 100 million of par claims but reserves worth USD 97 million after market repricing. Paying USD 30 million to redeeming holders leaves USD 67 million backing USD 70 million of claims, preserving the shortfall while reducing proportional coverage.
Common mistake: A stable market quotation does not prove that every holder can redeem at par promptly under stressed reserve-sale conditions.
Official syllabus reference: [1]
58. Tokenization and settlement finality
Tokenization represents claims or records on a digital infrastructure, but the token's legal and economic content depends on the underlying arrangement. Delivery-versus-payment design aims to link asset transfer with payment, reducing principal risk when both legs are reliably coordinated. It does not automatically remove operational, custody, smart-contract or legal-finality risk. Distinguish technical transaction confirmation from legally effective settlement and from access to the underlying asset. Interoperability across ledgers and conventional payment systems can reintroduce timing gaps. Faster settlement may also increase the need for prefunded liquidity rather than eliminate funding requirements.
Worked example: A tokenized bond transfers instantly on one ledger, but its cash payment settles later through a separate system. Unless the two legs are effectively linked, the seller can deliver the token before receiving final payment, retaining a principal exposure.
Common mistake: A technically confirmed token transfer does not by itself establish legal finality, valid ownership rights or simultaneous final payment.
Official syllabus reference: [1]
59. Private credit valuations and refinancing vulnerability
Private credit combines borrower credit risk with limited price discovery and often restricted exit liquidity. Floating-rate coupons can protect lender income from some rate movements while increasing borrower debt-service burdens. Refinancing risk rises when maturities arrive under tighter lending conditions or weaker valuations. Appraisal-based or model-based marks may adjust more slowly than public-market prices, making reported volatility look artificially low. Examine leverage, covenant terms, interest coverage and valuation assumptions together. Covenant protections can support intervention but do not guarantee recovery, and reported yield should be evaluated against expected losses, fees and liquidity constraints.
Worked example: A direct lender receives a higher floating coupon after rates rise, while the borrower's operating earnings remain flat. Interest coverage falls and refinancing becomes harder. The lender's higher current income therefore coexists with increased credit risk and possible future valuation pressure.
Common mistake: Higher floating-rate income does not automatically improve risk-adjusted performance when the same rate increase weakens borrowers' ability to service debt.
Official syllabus reference: [1]
60. Nonbank leverage and margin-driven contagion
Nonbank financial institutions can transmit stress through leverage, liquidity mismatch and common asset holdings. A market decline can trigger margin calls, forcing sales that depress prices and generate further calls across other investors. Initial losses may therefore be amplified by funding mechanics rather than fundamentals alone. Bilateral exposures can conceal aggregate leverage when each lender sees only part of a client's positions. Analyze gross financing, collateral reuse, redemption terms and concentrated counterparties alongside net market exposure. A market-neutral strategy can still create systemic liquidity demand if large offsetting positions require cash margin at different times.
Worked example: Several funds hold similar leveraged bond positions. A price decline raises their margin requirements, and each sells the same bonds to obtain cash. The resulting price pressure triggers additional calls, transmitting one market shock through shared financing and common holdings.
Common mistake: Low net directional exposure does not imply low liquidity risk when large gross positions generate substantial and uneven margin demands.
Official syllabus reference: [1]
Official syllabus sources
- FRM® Exam Information, Steps to Earn Certification | GARP
- FRM® Study Material, Guide, Books, Practice Exams | GARP