You’re on a roll
Create a free account to save your FRM progress, track your accuracy by topic, and pick up right where you left off.
Free Questions › FRM › Operational Risk and Resilience
Free FRM Operational Risk and Resilience Practice Questions & Answers
400 exam-style Operational Risk and Resilience questions. Pick your answer, hit Check answer, and see the worked solution — free to start, no signup.
100% free · No login to startQuestion 91
What is the 'procyclicality problem' in operational risk capital and how does Basel IV partially address it?
Select an option first.
Correct answer: B — Procyclicality: under Basel II BIA/SA, capital = f(gross income). During booms, gross income rises → capital rises (more buffer — but is it needed?). During downturns, gross income falls → capital falls (less buffer — exactly when OpRisk losses tend to be higher). This perverse dynamic reduces capital precisely when risk is elevated. Basel IV SA with ILM partially addresses this: capital now depends on both BI (forward-looking) and actual loss history (backward-looking) — if losses are rising, ILM > 1 increases capital
Explanation: B is correct. Procyclicality mitigation in Basel IV: BIA/SA flaw: falling gross income in a downturn directly reduces capital. Basel IV: ILM uses a 10-year rolling average loss — this smooths through the cycle (losses in booms and busts both contribute). Additionally, during a prolonged crisis with elevated losses, the ILM would increase, providing additional capital precisely when the loss environment is deteriorating. Limitation: the 10-year window smooths but doesn't fully eliminate cyclicality. A is wrong. C: procyclicality affects all risk types (market risk VaR is also procyclical).
Question 92
What is 'scenario weighting' in LDA and why is it sometimes applied?
Select an option first.
Correct answer: B — The practice of assigning reduced weight (less than 100%) to certain scenario estimates when combining them with internal loss data in the capital model — sometimes applied when scenarios are considered too conservative (anchored by a single catastrophic event that is judged to be idiosyncratic) or when the bank has made significant control improvements since the scenario was designed. Requires: governance approval, documentation of rationale, and regular review
Explanation: B is correct. Scenario weighting governance: under the AMA (now phased out), banks had discretion to weight scenarios in their capital models. This discretion was one of the AMA's weaknesses — scenarios could be underweighted to reduce capital. Under Basel IV SA: scenario analysis informs the ILM indirectly through loss data; it doesn't directly enter the formula. In ICAAP: scenario estimates may be weighted or adjusted with documented rationale. Key governance requirement: any weighting decision should be reviewed and approved by the OpRisk Committee or CRO, and documented for regulatory review. A is the opposite. C: weighting is an internal governance matter, not Basel IV SA.
Question 93
What is 'event classification' challenge in operational risk and what is the 'boundary event' problem?
Select an option first.
Correct answer: B — The difficulty of assigning a loss event to a single Basel event category when the event has characteristics of multiple categories — 'boundary events': (1) Unauthorised trading that is also a booking error (Internal Fraud + Execution/Delivery); (2) A system failure that enables external fraud (Technology Failure + External Fraud); (3) Mis-selling that also involved regulatory breach (Clients/Products + Compliance). Resolution: classify by the primary driver and document secondary classifications
Explanation: B is correct. Boundary events: the Basel Committee recognises that some losses fall between categories. Industry practice: use primary classification rule (which factor was the root cause?). Examples: rogue trading — primary = Internal Fraud (the deception was the core driver); the system failures that enabled it are secondary. IT failure enabling fraud — primary = External Fraud (the fraud was the ultimate driver) if the IT failure was just the vector. Documentation: capture both primary and secondary classifications in the loss database for analysis. A is about magnitude, not classification. C is a general data issue. D: classification challenges affect all banks.
Question 94
What is the 'insurance adjustment' to gross loss in operational risk and what is 'double-counting'?
Select an option first.
Correct answer: B — Subtracting insurance recoveries from gross losses to calculate net loss — but care must be taken to avoid double-counting: if insurance has already been reflected in reducing the gross loss estimate, it should not be subtracted again as a recovery. Also: future insurance recoveries (expected but not yet received) should not reduce current gross loss unless highly certain
Explanation: B is correct. Insurance adjustment accuracy: gross loss = full financial impact before insurance. Recovery = insurance payout received. Net loss = gross − recovery. Double-counting risk: (1) If the original loss estimate already assumes insurance will pay (i.e., it's estimated as net), then subtracting 'recovery' again understates the gross loss. (2) For ICAAP capital: using net losses in the LDA understates capital if insurance could become unavailable (insurer default, policy exclusion). Best practice: always record gross losses; record recoveries separately; use gross for capital modelling; net for management P&L reporting. A is a different concept. C: separate risk cell for insurance credit risk is a different concept. D is fraud.
Question 95
What is 'operational loss data capture threshold' and what is the typical industry standard?
Select an option first.
Correct answer: B — The minimum loss amount above which operational loss events must be captured in the internal loss database — below this threshold, losses are not recorded. Typical industry standard: €20,000 for large banks (consistent with ORX reporting threshold); some banks use €10,000 for more granular data. The threshold choice affects: capital model parameters (truncation adjustment needed), completeness of the loss database, and administrative cost (lower threshold = more events to record)
Explanation: B is correct. Capture threshold: the threshold is a practical compromise — recording every $10 processing error would overwhelm the database with low-value events that don't significantly affect capital. The €20K threshold (consistent with ORX data) ensures comparability with external data. Truncation correction: because losses below the threshold are not recorded, the statistical model must account for this truncation (the distribution is not observed below €20K). A: no single regulatory fixed threshold for all banks. C: banks set their own threshold but industry standards guide. D: applies to all event types, not just fraud.
Get the full FRM question bank — free
Drop your email and we'll send you fresh FRM practice questions, fully worked solutions and exam-deadline reminders. No spam — unsubscribe in one click.
Want to save your score and take a full mock exam? Create a free account →
Question 96
What is 'aggregation method' for combining risk cells and what are the main approaches?
Select an option first.
Correct answer: B — The method for combining operational risk capital estimates from individual risk cells to produce the total firm-wide operational risk capital — main approaches: (1) Simple summation (perfect correlation assumption — conservative upper bound); (2) Square root of sum of squares (independence assumption — provides maximum diversification); (3) Correlation matrix (intermediate — requires correlation estimates); (4) Copula-based aggregation (most sophisticated — captures tail dependence)
Explanation: B is correct. Aggregation approaches: Simple sum: $100M + $150M + $200M = $450M (assumes all cells fail simultaneously). Square root: √(100² + 150² + 200²) = √67,500 = $260M (assumes complete independence). Correlation matrix: intermediate result depending on assumed correlations (say, 0.5 average correlation → ~$340M). Copula: accounts for higher tail correlation than normal correlation suggests. Industry practice: most banks use correlation matrix or simple summation (conservative) for regulatory capital. Copula modelling is used in sophisticated internal models. A is summing with a fixed discount. C is a weighted average. D: Basel doesn't specify the aggregation method.
Question 97
What is 'coherent risk measure' and does the 99.9% VaR used for OpRisk satisfy this property?
Select an option first.
Correct answer: B — A risk measure is coherent (Artzner et al. 1999) if it satisfies: monotonicity, sub-additivity (ρ(A+B) ≤ ρ(A) + ρ(B)), positive homogeneity, and translational invariance. VaR at any confidence level is NOT sub-additive in general — the VaR of a combined position can exceed the sum of individual VaRs, violating the diversification principle. Expected Shortfall (ES/CVaR) IS a coherent risk measure. Despite this limitation, VaR at 99.9% is used for OpRisk Pillar 1 capital because of its simplicity and regulatory familiarity
Explanation: B is correct. VaR non-coherence: the sub-additivity failure of VaR is most pronounced for distributions with fat tails and discontinuities — common in operational risk. This means: the total OpRisk VaR across risk cells (using VaR aggregation) may overstate or understate diversification benefits. Expected Shortfall: satisfies all four coherence conditions. FRTB (market risk): shifted from VaR (99%) to ES (97.5%) partly for this reason. OpRisk: the regulatory standard remains VaR (99.9%) despite its theoretical limitations. Some advanced ICAAP models use ES. A: VaR is specifically not sub-additive. C: ES is coherent. D: VaR fails sub-additivity; ES satisfies all conditions.
Question 98
What does 'data imputation' mean in operational risk loss databases and when is it used?
Select an option first.
Correct answer: B — Filling in missing or incomplete data fields in the loss database using: (1) Interpolation (estimate from surrounding data points); (2) Mean/median substitution (replace missing value with average); (3) Model-based imputation (use regression or ML to predict the missing value from other fields); (4) Expert judgment (an expert estimates the missing amount). Used when: events were recorded with incomplete information, systems changed and historical data gaps exist, or data migration introduced gaps
Explanation: B is correct. Imputation in OpRisk: example — a loss event from 2010 is in the database but the recovery amount field is blank. Imputation: use the average recovery rate for that event type to estimate the recovery. Or: consult the original incident file. Imputation risk: imputed values introduce additional uncertainty into the model. Best practice: minimise the need for imputation through strong data governance (completeness requirements at entry). Document all imputed values and their method. Sensitivity analysis: test how sensitive the capital estimate is to alternative imputation assumptions. A is data deletion. C is inflation adjustment. D is a regulatory estimate requirement (different concept).
Question 99
What is 'right-tail risk' vs 'left-tail risk' and do they exist in operational risk?
Select an option first.
Correct answer: B — In operational risk, losses are strictly non-negative — the distribution has only a right tail (large positive losses). There is no 'left tail' of gains from operational risk events. This differs from market risk (which has both upside and downside). The right tail dominates capital requirements — the 99.9th percentile is entirely determined by the right tail of the aggregate loss distribution
Explanation: B is correct. OpRisk distributions: all operational losses are non-negative (you can't have a 'gain' from a fraud event in the direct loss sense). The distribution: most mass at small losses (many small process errors), thin right tail (rare catastrophic events like major rogue trading or cyber attacks). Capital: focused on the right tail at the 99.9th percentile. Compare to market risk: a trading portfolio can gain (right tail — profits) or lose (left tail — losses). VaR for market risk focuses on the left tail. A: wrong — only right tail. C: asymmetric in OpRisk (only losses). D: the tail has extreme but finite values in practice.
Question 100
What is 'stress period calibration' in operational risk and when is it used?
Select an option first.
Correct answer: B — Using historical data from a period of elevated operational losses to calibrate model parameters — ensures capital is not understated due to an unusually benign historical window. Under Basel (and analogous to stressed VaR in market risk), OpRisk stress calibration: identify a historical period with elevated operational losses relevant to the bank's profile; use that stressed window to test whether capital would be adequate during such a period
Explanation: B is correct. Stress calibration: OpRisk ICAAP stress testing: what would capital look like if we experienced a period like 2008-2010 (high conduct risk losses, elevated fraud)? Use historical stressed data to calibrate the ILM or to validate scenario estimates. Under Basel IV SA: the ILM uses the full 10-year window (which may include stressed periods). Economic capital models: some banks calibrate to a specifically defined stressed historical window to ensure capital robustness. A is too short a window. C is forward-looking scenario analysis. D is too recent (may not capture stressed periods).
Question 101
What is 'body-tail separation' in operational risk severity modelling?
Select an option first.
Correct answer: B — A modelling approach that fits different distributions to different parts of the severity distribution: a standard distribution (e.g., lognormal, Weibull) to the body of the distribution (frequent, moderate losses) and a heavy-tailed distribution (Pareto, GPD) to the tail (rare, severe losses above a threshold). This 'spliced distribution' provides better fit than a single distribution across the full range
Explanation: B is correct. Body-tail (spliced) distribution: f(x) = p × f_body(x) for x ≤ u, and (1-p) × f_tail(x-u) for x > u, where u is the splice point and p is the mixing weight. Advantages: lognormal fits small-to-medium losses well but underestimates the tail; Pareto/GPD models the tail accurately. The splice point u is typically chosen in the moderately high loss range (e.g., 90th percentile). The combined distribution provides a good fit across the full range — body fit for expected loss calculation, tail fit for capital estimation. A is a different technique. C is a different decomposition. D is event type separation.
Question 102
What does a 'fat tail' in the operational loss severity distribution imply for capital sensitivity?
Select an option first.
Correct answer: B — A fat-tailed distribution implies that capital is extremely sensitive to the estimated tail thickness (shape parameter ξ in GPD) — a small change in ξ (say from 0.5 to 0.7) can change the 99.9th percentile capital by 50-100%. This means: capital estimates have very wide confidence intervals, model risk is high, scenario analysis becomes critical for the tail, and excessive precision in reporting capital is misleading
Explanation: B is correct. Fat tail sensitivity: compare: Lognormal(μ=5, σ=1) → 99.9th percentile ≈ $1M. Pareto(ξ=0.8) → 99.9th percentile may be $5M. Change ξ to 0.9 → may jump to $15M. A 12.5% change in ξ triples the capital estimate. This is the 'model risk' embedded in OpRisk capital — high parameter uncertainty in heavy-tailed distributions creates enormous capital range. Best practice: report capital as a range (e.g., $50M-$200M) rather than a single number, reflecting parameter uncertainty. A: fat tails increase capital. C: normal distribution ≠ fat tail. D: fat tails are the primary driver of capital sensitivity.
Question 103
What is 'loss data consortium' and how does membership benefit individual banks?
Select an option first.
Correct answer: B — An industry collective where member banks share anonymised operational loss data — each member contributes their losses (above a threshold) and receives access to the pooled dataset and benchmarking reports. Benefits: access to tail loss data not available internally (rare large events at peer banks), industry benchmarks for capital validation, research and analytical support, and regulatory acceptance (regulators view consortium data as credible external data source)
Explanation: B is correct. Loss data consortia: ORX is the leading global consortium. Benefits in practice: a bank with 8 years of internal data and no losses > $50M gains access to ORX data showing 500+ events > $50M at member banks globally — dramatically improves tail estimation. Member obligations: data quality standards, consistent capture thresholds, accurate event classification. Confidentiality: data shared anonymously — banks can see industry losses but not which bank suffered them. A is supervisory data collection. C is a collective insurance pool (different). D: consortium data is not the same as the regulatory database.
Question 104
What is 'reverse stress testing' applied to operational risk?
Select an option first.
Correct answer: B — Identifying the specific operational risk scenarios or combination of events that would cause a bank to fail (breach a specified threshold) — working backwards from the failure outcome to determine what must happen to cause it. Reveals scenarios that forward stress tests might miss because: no one thought to specify them in advance. Required by FCA/PRA for UK banks; ICAAP supplement
Explanation: B is correct. Reverse stress testing for OpRisk: forward stress test: 'If a major cyber attack occurs, what is the financial impact?' Reverse stress test: 'What would cause us to breach our minimum capital requirement? What operational events, individually or in combination, would produce that outcome?' Example result: a simultaneous major regulatory fine AND a large system failure AND elevated fraud losses would deplete capital — prompting management to assess whether such combinations are plausible and whether resilience is adequate. A is sequential ordering. C is post-event analysis. D: required for significant firms; applicable at all sizes.
Question 105
What is the 'Exposure Indicator' concept in earlier OpRisk approaches and how was it used?
Select an option first.
Correct answer: B — In early operational risk capital proposals, exposure indicators (EIs) were measures of business activity used to proxy operational risk exposure — examples: transaction volume, assets under management, number of employees. The Basic Indicator Approach uses gross income as the exposure indicator. Different exposure indicators may better capture OpRisk for different business lines: retail banking (number of transactions), asset management (AUM), agency services (number of accounts)
Explanation: B is correct. Exposure indicators: the choice of exposure indicator significantly affects capital sensitivity to business changes. Gross income (BIA): if the bank grows revenue, capital increases even if risk management improves. Transaction volume: better for capturing processing risk (more transactions = more settlement errors). AUM: better for asset management operational risk. Limitations: no single indicator perfectly captures OpRisk for all business types. Basel IV's Business Indicator (BI) is a more refined version addressing some limitations of simple gross income. A is a credit risk concept. C is cyber-specific. D is event-level exposure.
Question 106
What does 'loss development' mean in the context of operational risk events?
Select an option first.
Correct answer: B — The tendency for the estimated final cost of an operational loss to change (usually increase) over time after initial discovery — an event discovered and initially estimated at $5M may develop to $50M as: legal costs accumulate, regulatory penalties are assessed, customer redress expands, reputational losses materialise. Insurance claims have similar 'reserve development' patterns. Loss databases should capture the 'as of' loss amount at each reporting date to track development
Explanation: B is correct. Loss development: initial booking: the event is discovered and an initial estimate is made. Development: as more information emerges, the estimate is revised. IBNR (Incurred But Not Reported): some losses have occurred but are not yet in the loss database. Final cost: may be determined years later (litigation, regulatory investigation). Capital implications: the LDA model uses finalised or latest-estimated losses — if losses are typically underestimated initially, the model may understate capital. Actuarial loss development methods (chain-ladder) are borrowed from insurance to project final losses from immature data. A is database growth. C: increases in capital may result from loss development. D is capital trend.
Question 107
What is 'Monte Carlo convergence' and how many simulations are needed in LDA?
Select an option first.
Correct answer: B — The property that Monte Carlo simulation results converge to stable values as the number of simulations increases — for LDA at the 99.9th percentile (a rare quantile), very many simulations are needed because only 0.1% of simulations fall in the region of interest. Typical: 100,000-1,000,000 simulations for stable 99.9th percentile estimates. Variance reduction techniques (importance sampling, stratified sampling) can reduce the required number
Explanation: B is correct. Convergence for rare quantiles: the 99.9th percentile: in 100,000 simulations, only 100 simulations fall in the top 0.1% — this is not many observations to estimate the quantile precisely. With 1,000,000 simulations: 1,000 observations in the tail — much more stable. Variance reduction: importance sampling oversamples the tail region, producing more stable tail estimates with fewer simulations. Practical compromise: 500,000-1,000,000 simulations with variance reduction. Stability test: increase simulations until the 99.9th percentile estimate stops changing. A: convergence depends on the quantile. C: 1,000 simulations is far too few for the 99.9th percentile. D: Basel doesn't set this.
Question 108
What is 'gross vs net income' in the context of the Business Indicator calculation?
Select an option first.
Correct answer: B — The Business Indicator uses absolute values and maxima of specific income components to prevent netting. For example, the Interest, Leases and Dividends Component uses |Net Interest Income| — the absolute value ensures that negative net interest income doesn't reduce the BI below zero. The Financial Component uses |Net P&L on trading book| — ensuring trading losses don't reduce BI. This prevents banks from using business losses to reduce their capital requirement
Explanation: B is correct. BI absolute values: BI = ILDC + SC + FC. Using absolute values and maxima: a bank with negative net trading income (trading losses) still has positive contribution to BI (absolute value of losses). This ensures BI captures the scale of activity, not just profitability. A bank that has large revenues and large expenses (high gross activity) will have a higher BI than one with small revenues and expenses, even if net income is the same. A: not always true. C: BI uses specific gross/absolute value components. D: they are different in the BI context.
Question 109
What is 'BCBS 355' and what did it reform in operational risk regulation?
Select an option first.
Correct answer: B — The consultation paper and final standard (issued 2016, incorporated into Basel IV) that reformed operational risk capital requirements — replacing the AMA with the Standardised Measurement Approach (SMA), subsequently refined and finalised as the new Standardised Approach (SA) in the Basel IV final standard (December 2017). Addressed: excessive variability in AMA capital, lack of comparability between banks, and model gaming concerns. The SA uses the Business Indicator and Internal Loss Multiplier
Explanation: B is correct. BCBS 355 (2016): consultation on the proposed Standardised Measurement Approach. Response: significant industry criticism (especially of the original ILM design). BCBS 356 (2016): final reform. Basel IV (December 2017): consolidated the SA into the broader Basel IV framework. Effective date: 2023 (US, EU, UK have varying implementation timelines). The reform was driven by: (1) AMA variability across banks; (2) Capital gaming through correlation assumptions; (3) Lack of comparability in Pillar 3 disclosures. A is credit risk. C is market risk (FRTB). D is LCR/NSFR.
Question 110
What is 'estimation bias' in scenario analysis and what causes it?
Select an option first.
Correct answer: B — Systematic errors in expert probability and impact estimates — caused by: (1) Availability bias (over-weighting recent or vivid events); (2) Overconfidence (confidence intervals too narrow); (3) Anchoring (initial estimates anchor subsequent revisions); (4) Motivational bias (business managers want low capital so underestimate scenarios); (5) Group composition bias (homogeneous groups reach consensus too quickly). Requires structured facilitation and calibration against external benchmarks to reduce
Explanation: B is correct. Debiasing scenario estimates: (1) Decomposition: break the scenario into components (frequency × impact) and estimate each separately before combining. (2) Reference classes: use the full range of similar events globally, not just the most memorable. (3) Pre-mortem: 'imagine this event has occurred — what sequence of events led here?' (4) External benchmarks: share ORX or public loss data after initial estimates to calibrate. (5) Independent estimates: collect estimates individually before group discussion to prevent anchoring on the first speaker. A is deliberate manipulation (a separate issue). C is statistical sampling error. D: bias exists in internal scenario estimates too.
Question 111
What is 'validation sampling' in operational risk loss data quality assessment?
Select an option first.
Correct answer: B — Testing the accuracy and completeness of the loss database by sampling a subset of entries and verifying them against source documents (accounting records, incident reports, insurance claims, legal files) — validates: loss amounts (are they correct?), classifications (right event type and business line?), dates (accounting and discovery dates correct?), completeness (are all qualifying events in the database?)
Explanation: B is correct. Validation sampling: internal audit and model validation teams conduct regular sampling exercises. Sample size: statistical approach to determine how many samples are needed to give confidence in the full database. Key findings: loss amounts often under-reported (conservative bias); classifications frequently inconsistent (same type of event mapped to different categories by different business units); completeness gaps (business units with low loss reporting rates may be failing to capture all qualifying events). A is scenario selection. C is a different concept. D is the regulator's role.
Question 112
What is 'risk cell granularity' and what are the tradeoffs?
Select an option first.
Correct answer: B — Finer granularity (more risk cells) provides better risk differentiation but suffers from data sparsity within each cell. Coarser granularity has more data per cell but may mask important risk differences. Optimal granularity balances: data sufficiency (enough observations per cell for stable estimation), meaningful risk differentiation (cells should represent genuinely different risk profiles), and operational feasibility (managing hundreds of cells is impractical)
Explanation: B is correct. Granularity tradeoff: a bank with 100 loss events per year split across 56 cells averages under 2 events per cell — far too few for stable distribution fitting. Better practice: aggregate cells with similar risk profiles and supplement sparse cells with external data. The appropriate granularity depends on the bank's size, complexity, and available data. Large banks: more granularity is feasible. Small banks: fewer cells, more data pooling. A: data sparsity reduces quality for overly granular cells. C: the 56-cell matrix is a maximum, not a requirement. D: granularity should reflect the bank's actual risk profile.
Question 113
What is the 'loss frequency rate' (LFR) and how is it used as a KRI?
Select an option first.
Correct answer: B — The number of operational loss events per unit of business activity (e.g., loss events per 10,000 transactions, per employee, or per year) — used as a KRI to monitor whether the rate of operational incidents is changing over time, independently of individual loss severity. A rising LFR indicates increasing frequency even if no single event has been large
Explanation: B is correct. Loss frequency rate: a stable or declining LFR is a positive signal (controls improving). A rising LFR may indicate: deteriorating processes, new activities without adequate controls, or a change in the external threat environment. Monitoring: track LFR by event type and business unit — which categories are driving the change? LFR complements severity monitoring. A bank with stable LFR but increasing average severity has a different risk profile than one with rising LFR. A is a different loss metric. C is a fraud mix ratio. D is a near-miss rate.
Question 114
What is 'Peaks-over-Threshold' vs 'block maxima' in extreme value theory?
Select an option first.
Correct answer: B — Block maxima: divide the data into blocks (e.g., annual); take the maximum of each block; fit a Generalised Extreme Value (GEV) distribution to the maxima. Peaks-over-Threshold (POT): fit a Generalised Pareto Distribution (GPD) to all exceedances above a threshold. POT is preferred for operational risk: it uses more of the available extreme data (all large losses, not just the annual maximum), reducing estimation uncertainty
Explanation: B is correct. POT advantages: with 10 years of annual data, block maxima gives 10 data points (annual maxima). POT with a high threshold (90th percentile) might give 100+ data points — much more data for fitting the GPD. More data → more precise parameter estimates → more reliable capital estimates. Block maxima advantage: theoretically pure (GEV convergence theorem), but practically limited by small sample sizes. For OpRisk: POT/GPD is the dominant approach. A: different methods produce different results. C: block maxima preferred in some contexts (e.g., insurance). D: different distributions (GEV vs GPD).
Question 115
What is 'inter-arrival time' in operational risk modelling and what distribution models it?
Select an option first.
Correct answer: B — The time between successive operational loss events — for a Poisson process, inter-arrival times follow an exponential distribution with mean = 1/λ. This provides an alternative way to model frequency: instead of modelling the count N per year (Poisson), model the time between events (exponential). Both are equivalent representations of the same process
Explanation: B is correct. Inter-arrival times: for a homogeneous Poisson process with rate λ: number of events per period ~ Poisson(λ); inter-arrival times ~ Exponential(λ). This duality is useful for simulation: instead of drawing N from Poisson and then N severity draws, simulate successive inter-arrival times until the total exceeds one year, drawing severity for each event. Equal result. Practical relevance: testing whether observed inter-arrival times follow exponential distribution is equivalent to testing the Poisson assumption. A: exponential, not lognormal. C is a process latency measure. D is a regulatory cycle.
Question 116
What is 'skewness' of the aggregate loss distribution in LDA and what drives it?
Select an option first.
Correct answer: B — The aggregate loss distribution in LDA is highly positively skewed — most simulated years have low total losses, but occasional simulations produce very large totals due to rare high-severity events. Driven by: (1) Heavy-tailed severity distribution (occasional catastrophic events); (2) Low-frequency events (more Poisson variability when λ is small); (3) Positive skewness of the severity distribution. This positive skewness is why the 99.9th percentile capital can be many multiples of the expected loss
Explanation: B is correct. Positive skewness consequences: the mean (expected loss) is far below the 99.9th percentile capital. Example: mean aggregate loss = $5M/year; 99.9th percentile = $500M. The ratio (100:1) reflects the extreme skewness. This skewness means: normal distribution assumptions dramatically underestimate capital; standard mean-variance approaches are inappropriate for OpRisk. A: highly positively skewed. C: equity returns can be negatively skewed. D: skewness determines how far the 99.9th percentile exceeds the mean — very relevant for capital.
Question 117
What is 'correlated scenarios' in operational risk and how does co-occurrence affect capital?
Select an option first.
Correct answer: B — When multiple severe operational risk scenarios could occur simultaneously — e.g., a major cyber attack triggering a business disruption event, which simultaneously causes a client impact event. If capital is calculated for each scenario independently and then summed, this may double-count if the scenarios are actually caused by the same underlying event. Capital should reflect the realistic joint probability
Explanation: B is correct. Co-occurrence and capital: independent scenarios: add capitals (upper bound). Fully correlated scenarios (same event causes multiple losses): use the single larger capital estimate. Partial correlation: weighted approach. In practice: a single cyber event triggers cascades — data breach (external fraud), system downtime (business disruption), customer harm (clients/products). The total loss from the event must be captured as a single scenario, not three separate ones. This avoids both double-counting and under-counting. A: shared root cause is one source of correlation but not the only definition. C: correlation reduces capital through diversification only if scenarios are genuinely independent. D: regulatory models should capture correlations.
Question 118
What is 'the role of actuaries' in operational risk management?
Select an option first.
Correct answer: B — Actuaries bring quantitative skills in: (1) Loss distribution fitting (maximum likelihood estimation, goodness-of-fit testing); (2) Credibility weighting (blending internal and external data); (3) Extreme value theory (GPD, POT methodology); (4) Aggregate loss distribution modelling (compound frequency-severity, Panjer recursion); (5) Insurance pricing and risk transfer valuation; (6) Scenario analysis calibration. Their casualty actuarial science background provides the statistical toolkit directly applicable to operational risk capital modelling
Explanation: B is correct. Actuarial skills in OpRisk: the LDA framework is borrowed directly from casualty actuarial science (collective risk model). Major banks employ actuaries in their operational risk capital teams for the quantitative aspects. Actuarial standards: the Casualty Actuarial Society (CAS) and Institute and Faculty of Actuaries (IFoA) have developed guidance on operational risk applications of actuarial methods. A: actuaries work in banking OpRisk, not only insurance. C: actuarial methods are foundational. D: data scientists complement actuaries but don't replace the core statistical framework.
Question 119
What is 'parameter instability' in LDA models over time and how is it addressed?
Select an option first.
Correct answer: B — The fitted parameters of frequency and severity distributions may change over time as: the loss history evolves (new large events enter the dataset), control improvements reduce loss frequency, business mix changes, or the external threat environment evolves. Instability: capital estimates may jump significantly when a large historical event enters or exits the rolling data window (known as the 'cliff edge' effect)
Explanation: B is correct. Addressing parameter instability: monitoring: track parameter estimates at each capital model update — significant changes require investigation and communication. Data windowing: a 10-year rolling window means a large event from 10 years ago eventually drops out, potentially reducing capital sharply (cliff edge). Solutions: (1) Use longer windows to smooth parameter changes; (2) Weight recent data more heavily (but gradually); (3) Add a stability buffer to capital estimates; (4) Scenario analysis for the tail ensures capital doesn't drop sharply when historical tail events exit. A: parameters evolve with the data. C: instability can increase or decrease capital. D: affects all heavy-tailed distributions.
Question 120
What is 'validation frequency' for operational risk capital models?
Select an option first.
Correct answer: B — The frequency at which independent model validation is conducted — typically: annual comprehensive validation (full review of data quality, methodology, assumptions, backtesting), triggered re-validation when material changes occur (significant new loss events, model redevelopment, business changes, regulatory change), and quarterly performance monitoring (KRI tracking, ongoing backtesting). Annual validation aligned with ICAAP submission cycle
Explanation: B is correct. Validation cycle: annual: full independent validation — conceptual soundness, data quality, methodology review, sensitivity analysis. Triggered: material change in the loss profile (a $500M loss event enters the database), major business restructuring, new products substantially changing the risk profile, or regulatory feedback requiring changes. Ongoing: monitoring by the model owner — are frequency and severity distributions still fitting the updated data? Performance reporting. MRM policy: model risk policy should define validation frequency by materiality tier. A: proactive schedule, not reactive. C: regulatory requirements exist. D: frequency calibrated to model importance and materiality.
More free FRM topics
Ten questions in
- The ones you miss are saved as a drill you can repeat
- Your place is kept, on this device and any other
- A streak, if that is the thing that gets you back tomorrow
Every question on this page stays free and open either way.