This free guide connects financial risk decisions with the calculations behind them: portfolio diversification, statistical inference, derivatives, valuation, and loss measurement. Twenty original multiple-choice questions provide a representative preview, with worked explanations and a rationale for every option. Sixty distinct lessons explain mechanisms, practical examples, and common errors. Each of the four 2026 Part I domains receives five questions and fifteen concepts; this exercise allocation is not an official weighting. The supplied GARP source verifies domain relevance, not every technical statement or detailed reading. Scope is the 2026 curriculum as of October 16, 2026; future announcements do not establish a replacement syllabus.
What this guide covers
Explore 20 original FRM Part I questions and 60 worked concepts across the four 2026 risk domains.
Foundations of Risk Management
1. Risk taxonomy and transmission channels
Market risk arises from changes in prices, rates, spreads, or volatility; credit risk arises from deterioration or default of an obligor. Funding liquidity risk concerns obtaining cash when needed, while market liquidity risk concerns trading without excessive price impact. Operational risk concerns failures involving processes, people, systems, or external events. These categories interact rather than form mutually exclusive boxes. A useful classification identifies the initiating event, the exposure transmitting it, and the resulting loss. Classification supports ownership and measurement without implying that one incident has only one risk type.
Worked example: A dealer suffers a bond price decline, receives a collateral call, and sells other assets into a thin market. The initial market loss creates funding pressure, which then produces additional losses through market liquidity.
Common mistake: Assigning the entire episode to market risk hides the funding and liquidation mechanisms that can amplify the original loss.
Official syllabus reference: [2]
2. Risk appetite, capacity, and limits
Risk capacity is the amount of risk an organization can bear given capital, liquidity, obligations, and other constraints. Risk appetite expresses the risk it chooses to accept in pursuing objectives. Limits translate that preference into actionable restrictions such as exposure ceilings, concentration caps, or stress-loss budgets. A limit should identify its metric, horizon, owner, and escalation process. Appetite can be below capacity because preserving flexibility has value. A numerical limit does not fully express appetite unless its assumptions and relationship to adverse outcomes are understood.
Worked example: A lender can withstand a modeled USD 80 million stress loss but chooses an appetite of USD 50 million. It assigns sector limits whose combined stressed exposures remain consistent with that chosen tolerance and available diversification.
Common mistake: Treating maximum financial capacity as the desired operating target ignores the organization's chosen buffer and uncertainty around measured losses.
Official syllabus reference: [2]
3. Independent challenge and risk ownership
Business managers own the risks created by their activities, while an independent risk function measures exposures, challenges assumptions, and escalates breaches. The board establishes oversight and approves material aspects of risk appetite. Internal audit supplies independent assurance about the effectiveness of governance and controls. Clear responsibilities reduce the chance that revenue incentives suppress unfavorable information. Independence requires credible reporting access and authority, not merely a separate job title. Governance should also specify who can approve exceptions, how long exceptions last, and what remediation is required.
Worked example: A desk exceeds its concentration limit after a market move. The desk proposes a reduction schedule, the risk function evaluates interim exposure, and the authorized committee decides whether a temporary exception fits the firm's appetite.
Common mistake: Making the independent risk function solely responsible for trading losses can weaken business accountability and blur the distinction between oversight and ownership.
Official syllabus reference: [2]
4. Enterprise aggregation and dependence
Enterprise risk aggregation combines exposures across businesses while recognizing dependence, offsets, and concentration. Adding standalone risk measures can overstate diversified risk, but assuming stable low correlations can understate common shocks. Dependencies may be nonlinear or become stronger during stress. Offsets also require compatible horizons and realizable economic relationships; a long-term gain may not finance an immediate collateral call. Effective aggregation therefore pairs statistical measures with factor concentrations and scenarios. Its objective is to describe the firm's combined vulnerability rather than simply total departmental reporting numbers.
Worked example: Two desks appear diversified because one holds corporate loans and the other sells equity protection. A recession scenario reveals that both lose when corporate earnings weaken, so their different product labels conceal shared economic exposure.
Common mistake: Assuming different business names imply diversification overlooks common borrowers, collateral, funding sources, and macroeconomic drivers across the enterprise.
Official syllabus reference: [2]
5. CAPM and the market risk premium
The capital asset pricing model gives expected return as the risk-free rate plus beta times the expected market risk premium. Beta measures covariance with the market divided by market variance. Under the model's assumptions, diversified investors require compensation for systematic exposure rather than diversifiable standalone volatility. The security market line plots expected return against beta and can describe individual securities or portfolios. An observed return above the line does not automatically establish mispricing because expected returns and betas are estimated, and the model may omit relevant risks.
Worked example: With a risk-free rate of 2%, expected market return of 7%, and beta of 1.4, a stock's CAPM expected return is 9%. The calculation adds a 7% systematic premium to the risk-free return.
Common mistake: Multiplying beta by the full market return overstates the premium because the risk-free component should not receive the beta multiplier.
Official syllabus reference: [2]
6. Beta versus total return volatility
Beta measures sensitivity to market returns, whereas volatility measures dispersion of a security's own returns. Their relationship is beta = correlation with the market times security volatility divided by market volatility. A highly volatile security can have modest beta if much of its variation is unrelated to the market. In a single-factor regression, total variance separates into systematic variance and residual variance when the residual is uncorrelated with the market. Diversification reduces residual risk across sufficiently independent holdings but leaves exposure to the common market factor.
Worked example: A stock has volatility of 30%, market volatility is 20%, and their correlation is 0.40. Its beta is 0.60, showing that substantial standalone volatility can coexist with below-market sensitivity in a diversified portfolio.
Common mistake: Ranking securities by volatility as though that also ranks beta ignores how strongly each security moves with the market.
Official syllabus reference: [2]
7. Efficient frontier and capital market line
The mean-variance efficient frontier contains portfolios offering the highest expected return for a given variance, or lowest variance for a given expected return. With a risk-free asset and common borrowing and lending assumptions, the capital market line joins the risk-free rate to the tangency portfolio. Its slope is the tangency portfolio's Sharpe ratio. The capital market line describes efficient combinations, while the security market line relates expected return to beta. Real financing constraints, estimation error, and nonnormal preferences can limit the practical relevance of the idealized frontier.
Worked example: A tangency portfolio has expected return of 8% and volatility of 12%, while cash earns 2%. Investing half in each gives expected return of 5% and volatility of 6%, assuming cash has zero volatility.
Common mistake: Placing every individual security on the capital market line confuses efficient portfolio combinations with the separate beta-based pricing relationship.
Official syllabus reference: [2]
8. Arbitrage pricing and factor exposures
A multifactor return model separates exposure to common economic drivers from asset-specific variation. In an arbitrage pricing framework, expected excess return is a linear combination of factor sensitivities and factor risk premiums under the relevant assumptions. Factor realizations explain unexpected returns; factor premiums describe compensation for bearing exposure. The distinction prevents using an observed inflation surprise as though it were an expected premium. A negative loading can lower required return when it provides protection against a positively priced factor. Model interpretation depends on the factors' definitions and normalization.
Worked example: Suppose the risk-free rate is 2%, two factor premiums are 4% and 1%, and a security's loadings are 0.8 and −0.5. Its model-implied expected return is 4.7%, with the negative loading reducing compensation.
Common mistake: Adding factor loadings directly to expected return ignores the premiums attached to those exposures and can mix incompatible units.
Official syllabus reference: [2]
9. Sharpe and Treynor performance measures
The Sharpe ratio divides excess return by total return volatility. The Treynor measure divides excess return by market beta. Sharpe is useful when evaluating the entire risky investment or when residual risk matters to the investor. Treynor focuses on systematic exposure and is more interpretable for well-diversified holdings within a broader portfolio. Rankings can differ because the denominators measure different risks. Returns, risk-free rates, and risk estimates must use compatible periods. Negative excess returns or near-zero beta require special care because simple ratio rankings can become misleading.
Worked example: A fund earns 10% when cash earns 2%, with volatility of 16% and beta of 0.8. Its Sharpe ratio is 0.50 and Treynor measure is 10%, each describing performance against a different risk denominator.
Common mistake: Choosing Treynor for a concentrated standalone investment can overlook material residual risk that the investor has not diversified away.
Official syllabus reference: [2]
10. Jensen's alpha and the information ratio
Jensen's alpha measures return beyond a specified asset-pricing model's expected return, commonly CAPM. It asks whether performance exceeds compensation for modeled systematic exposure. The information ratio divides average active return relative to a chosen benchmark by tracking error, the volatility of active returns. It evaluates consistency of benchmark-relative performance rather than total portfolio risk. Both measures depend on benchmark or model selection. An apparent alpha can disappear when omitted risk factors are included, while a high information ratio against an unsuitable benchmark may have little investment meaning.
Worked example: A portfolio earns 9%, while its CAPM expected return is 8%, producing alpha of 1%. Against a separate benchmark it averages 2% active return with 4% tracking error, producing an information ratio of 0.50.
Common mistake: Treating alpha and active return as interchangeable ignores that one uses a risk model while the other uses a chosen benchmark.
Official syllabus reference: [2]
11. RAROC and unexpected-loss capital
Risk-adjusted return on capital compares a defined measure of risk-adjusted earnings with capital allocated to absorb adverse outcomes. A common numerator subtracts operating expenses and expected loss from revenue, while economic capital addresses unexpected loss over a chosen horizon and confidence convention. Expected loss and capital therefore serve different purposes. RAROC helps compare businesses with different risk intensities, provided definitions are consistent. Hurdle comparisons should reflect capital opportunity cost. Ratio optimization alone can miss differences in scale, incremental capital, strategic constraints, and interactions between existing and proposed exposures.
Worked example: A business earns USD 18 million revenue, incurs USD 10 million expenses and USD 2 million expected loss, and requires USD 30 million economic capital. Under this definition, its risk-adjusted income is USD 6 million and RAROC is 20%.
Common mistake: Subtracting economic capital as though it were an annual expense confuses a loss-absorbing resource with the cost of supplying that resource.
Official syllabus reference: [2]
12. Why corporate hedging can create value
In ideal frictionless markets, diversified shareholders can often adjust financial exposures themselves. Corporate hedging can create value when real-world frictions make cash-flow volatility costly. Relevant mechanisms include financial distress costs, expensive external financing, underinvestment after adverse shocks, and nonlinear tax effects. A hedge should address a specified exposure and objective rather than maximize derivative profits. Reducing accounting earnings volatility is not always equivalent to reducing economic risk. Basis mismatch, collateral demands, and counterparty exposure can offset some benefits and should be included in the economic assessment.
Worked example: An exporter expects foreign-currency receipts needed to finance a factory project. Selling that currency forward stabilizes domestic funding, reducing the chance that an adverse exchange-rate move forces expensive borrowing or delays an otherwise valuable investment.
Common mistake: Judging a hedge solely by its derivative profit ignores the offsetting cash-flow exposure that motivated the hedge in the first place.
Official syllabus reference: [2]
13. Model risk and effective validation
Model risk arises when a model's design, implementation, inputs, or use produces misleading decisions. Validation examines conceptual soundness, data suitability, coding, calibration, outcomes, and limitations. Backtesting supplies evidence about performance but cannot prove reliability outside observed conditions. A model can be implemented correctly yet answer the wrong business question. Effective challenge therefore tests assumptions and intended use as well as numerical accuracy. Controls may include benchmark models, sensitivity analysis, documented restrictions, and independent review. Material changes in markets or application can require renewed validation even without code changes.
Worked example: A pricing model reproduces historical prices accurately but assumes continuous liquidity. Before using it to value a concentrated position during stress, validators examine liquidation assumptions and compare values under wider spreads and discontinuous trading conditions.
Common mistake: Equating a clean software test with complete validation overlooks whether the model's economic assumptions remain appropriate for its intended decision.
Official syllabus reference: [2]
14. Leverage and liquidity feedback spirals
A liquidity spiral occurs when initial losses trigger funding demands or leverage constraints, forcing asset sales that depress prices and generate further losses. Higher haircuts can require more collateral even when a position's fundamental cash flows have not changed. Mark-to-market losses then reduce equity, potentially forcing additional deleveraging. Market liquidity and funding liquidity reinforce each other through this feedback. Historical crisis analysis is useful for tracing these mechanisms rather than memorizing isolated dates. A position can be solvent on a long-term valuation basis yet fail to meet immediate cash obligations.
Worked example: A fund finances USD 100 million of assets with USD 95 million borrowing. A price decline reduces equity, and its lender simultaneously raises the haircut. Forced sales into a weak market deepen losses and collateral needs.
Common mistake: Assuming a low long-run default probability guarantees survival ignores the possibility that short-term funding demands force liquidation before recovery.
Official syllabus reference: [2]
15. Professional integrity in risk reporting
Ethical risk reporting requires accurate representation of results, assumptions, limitations, and material conflicts. Analysts should distinguish observed data from estimates and avoid presenting uncertain model outputs as facts. Confidential information requires appropriate handling, while conflicts can affect model selection, valuation judgments, and communication. An analytically correct calculation can still mislead if unfavorable scenarios or qualifying assumptions are deliberately omitted. Practical integrity includes preserving a traceable basis for conclusions and communicating material uncertainty to decision-makers. These principles explain professional judgment without asserting particular disciplinary procedures or jurisdiction-specific legal duties.
Worked example: A risk analyst discovers that a proposed calibration lowers reported exposure mainly by excluding volatile months. The analyst reports both calibrations, explains the exclusion's effect, and discloses the business sponsor's preference before the approval decision.
Common mistake: Believing disclosure is unnecessary because every displayed number is mathematically correct overlooks how selective assumptions can materially mislead the report's users.
Official syllabus reference: [2]
Quantitative Analysis
16. Conditional probability and Bayes' rule
Conditional probability restricts the reference population: P(A|B) = P(A and B)/P(B), provided P(B) is positive. Bayes' rule reverses a conditioning relationship by combining a likelihood with a prior probability. The denominator accounts for every way the observed evidence can occur. This is crucial when an event is rare because false positives among the much larger unaffected population can dominate observations. Independence means conditioning does not change probability, but mutually exclusive positive-probability events are dependent. A probability tree or frequency representation often makes the denominator transparent.
Worked example: Among 10,000 loans, suppose 100 default. An alert catches 80 of those but also flags 990 surviving loans. Among 1,070 flagged loans, only 80 default, giving a posterior probability of approximately 7.48%.
Common mistake: Reading an alert's sensitivity as the probability of default after an alert reverses the conditioning and ignores the underlying default rate.
Official syllabus reference: [2]
17. Expectation, covariance, and portfolio variance
Expectation is linear, so a weighted portfolio's expected return is the weighted sum of component expected returns regardless of dependence. Variance is not linear because co-movement matters. For two assets, portfolio variance is w1²sigma1² + w2²sigma2² + 2w1w2Cov(R1,R2). Covariance equals correlation times the two standard deviations. Positive covariance reduces diversification benefits, while negative covariance can provide substantial offsets. The formulas describe the specified return distribution and weights; changes in weights, nonlinear payoffs, or unstable dependence require further modeling rather than mechanical reuse of historical estimates.
Worked example: Two equally weighted assets each have volatility of 20% and zero correlation. Portfolio variance is 0.25 times 0.04 plus 0.25 times 0.04, or 0.02. Portfolio volatility is approximately 14.14%, below either asset's volatility.
Common mistake: Averaging component volatilities as though volatility were linear discards covariance and usually misstates the actual risk of the combined portfolio.
Official syllabus reference: [2]
18. Normal returns and lognormal prices
A normally distributed variable can take any real value, whereas a lognormal variable is strictly positive. If continuously compounded return is normal, the corresponding gross price ratio is lognormal. For log return X with mean m and variance s², E(exp(X)) = exp(m + s²/2), so exponentiating the mean alone understates the expected gross return. Simple and log returns are close for small changes but differ materially over larger moves. A lognormal price assumption excludes negative prices and does not automatically represent jumps, volatility clustering, or heavy empirical tails.
Worked example: A one-year log return has mean 0.04 and variance 0.09. Expected gross price growth is exp(0.085), approximately 1.0887. Exponentiating only 0.04 would give 1.0408 and miss the effect of return dispersion on expected price.
Common mistake: Assuming normal log returns make the future price normally distributed confuses the modeled logarithm with the positive price obtained by exponentiation.
Official syllabus reference: [2]
19. Skewness, kurtosis, and tail shape
Skewness measures asymmetry using the standardized third central moment. Negative skewness indicates a distribution with relatively pronounced downside outcomes, though it does not fully describe every tail feature. Kurtosis uses the standardized fourth central moment and is highly sensitive to extreme observations. Normal kurtosis is three, or zero when reported as excess kurtosis. Two distributions with the same mean and variance can have very different loss probabilities. Moment estimates themselves can be unstable in short samples, so a seemingly precise kurtosis estimate should not substitute for inspecting tail observations and model assumptions.
Worked example: Two strategies both average 1% monthly return with 3% volatility. One has occasional large losses offset by frequent small gains. Its negative skewness signals a materially different experience despite matching the other's first two moments.
Common mistake: Treating equal volatility as equal downside risk ignores asymmetry and tail thickness, which can strongly affect severe-loss probabilities.
Official syllabus reference: [2]
20. Sampling distributions and the central limit theorem
A sampling distribution describes how a statistic varies across repeated samples. For independent, identically distributed observations with finite variance, the sample mean has variance sigma²/n. The central limit theorem gives an approximately normal standardized sample mean as sample size grows under its conditions. It does not make the original observations normal or guarantee accurate approximation in every finite sample. Dependence can change effective sample size, and extremely heavy tails may invalidate familiar variance-based results. Standard error measures uncertainty in an estimator rather than dispersion of individual observations.
Worked example: Daily returns have standard deviation of 2%. With 400 independent observations, the sample mean's standard error is 0.10%. Collecting four times as many comparable observations halves that standard error rather than reducing it to one quarter.
Common mistake: Using the central limit theorem to declare individual returns normal confuses the distribution of observations with the distribution of their sample mean.
Official syllabus reference: [2]
21. Confidence intervals and precision
A confidence interval combines an estimate with a margin based on its standard error and a reference distribution. For a normal population mean with known variance, the interval uses a normal critical value; with estimated variance under the usual normal assumptions, a Student t critical value accounts for additional uncertainty. Confidence describes the long-run coverage of the procedure, not a probability assigned to a fixed parameter after observing one interval. Wider confidence levels increase interval width. Larger independent samples improve precision, but they do not automatically remove selection bias or model misspecification.
Worked example: A sample mean loss is USD 20 with standard error USD 3. Using a supplied critical value of two gives an interval from USD 14 to USD 26. Doubling precision requires reducing standard error, not changing the estimate.
Common mistake: Calling a narrow interval accurate without examining sampling bias confuses precision around an estimate with proximity to the correct population quantity.
Official syllabus reference: [2]
22. Hypothesis tests, errors, and power
A hypothesis test compares a statistic with its distribution under a specified null hypothesis. A Type I error rejects a true null; a Type II error fails to reject a false null. Power is one minus the Type II error probability at a specified alternative. Holding sample size fixed, a stricter significance threshold generally reduces false rejections but can reduce power. A p-value measures extremeness under the null, not the probability that the null is true. Statistical significance also differs from economic importance, especially in large samples with very precise estimates.
Worked example: A bank tests whether a model's mean pricing error is zero. With many observations, an average error of one cent may be statistically significant while remaining economically small relative to trading costs and transaction sizes.
Common mistake: Interpreting a p-value of 0.03 as a 3% probability that the null hypothesis is true reverses its conditional meaning.
Official syllabus reference: [2]
23. OLS coefficients and prediction
Ordinary least squares chooses coefficients to minimize the sum of squared residuals. In a simple regression with an intercept, the slope equals sample covariance between X and Y divided by sample variance of X. The intercept aligns the fitted line with sample means. A slope describes an association conditional on the model; causal interpretation requires stronger identification assumptions. Predictions also depend on whether the new observation resembles the estimation sample. A confidence interval for the conditional mean is narrower than a prediction interval for an individual outcome because the latter includes unexplained outcome variation.
Worked example: A fitted cost model is cost = 100 + 4 × volume, with dollars and units as defined. At volume 30, predicted cost is USD 220. Actual cost can differ because the regression leaves residual variation unexplained.
Common mistake: Reading the fitted slope as a proven causal effect ignores omitted variables, reverse causality, and other threats to the model's interpretation.
Official syllabus reference: [2]
24. Heteroskedasticity and serial correlation
Heteroskedasticity means error variance differs across observations; serial correlation means errors are correlated across time. Under suitable exogeneity conditions, OLS coefficients can remain unbiased or consistent, but conventional standard errors may be wrong and OLS may lose efficiency. Heteroskedasticity-robust standard errors address changing variance, while appropriate heteroskedasticity-and-autocorrelation-consistent methods also address temporal dependence. These corrections do not repair endogeneity or a wrong conditional-mean specification. Residual plots and formal diagnostics help identify the issue, but the remedy must match the dependence structure and intended inference.
Worked example: A return regression shows small residuals during calm months and large residuals during crises. Robust standard errors can adjust inference for unequal variance, but they do not add missing crisis-related explanatory variables to the fitted mean equation.
Common mistake: Assuming robust standard errors cure every regression problem overlooks that biased coefficients from endogeneity remain biased after changing the estimated uncertainty.
Official syllabus reference: [2]
25. Omitted variables and multicollinearity
Omitting a relevant explanatory variable biases a retained coefficient when the omitted variable affects the outcome and correlates with the retained regressor. In the simple two-regressor case, the bias equals the omitted coefficient times Cov(X,Z)/Var(X). Multicollinearity is different: highly related included regressors make individual coefficients imprecise, even when the model is correctly specified. Perfect collinearity prevents unique coefficient estimation. Adding variables can reduce omitted-variable bias while increasing sampling uncertainty. The decision should follow the economic question and identification assumptions rather than maximizing fit mechanically.
Worked example: A model relates bond spreads to leverage but omits profitability. If profitability reduces spreads and is negatively correlated with leverage, the omitted contribution biases leverage's estimated coefficient upward, combining leverage effects with the omitted profitability relationship.
Common mistake: Treating multicollinearity as the same problem as omitted-variable bias confuses imprecise estimation with systematic distortion of the target coefficient.
Official syllabus reference: [2]
26. Stationarity and AR(1) mean reversion
For an AR(1) process Xt = c + phi × X(t−1) + error, covariance stationarity requires absolute phi below one under the usual innovation assumptions. Its unconditional mean is c/(1 − phi). Shocks decay geometrically at rate phi, so a coefficient near one implies slow mean reversion. At phi equal to one, a unit-root process does not have the same finite stationary variance. Regressing unrelated persistent level series can produce misleading significance. Differencing can remove a unit root, but it changes the modeled quantity and may discard useful long-run relationships.
Worked example: A spread follows Xt = 0.4 + 0.8X(t−1) plus a zero-mean shock. Its long-run mean is two. If today's spread is three, next period's conditional expectation is 2.8, moving partway toward that mean.
Common mistake: Interpreting a coefficient near one as rapid adjustment reverses the mechanism: persistent shocks decay slowly when the autoregressive coefficient is large.
Official syllabus reference: [2]
27. EWMA and GARCH volatility dynamics
EWMA updates conditional variance by combining the previous variance with the latest squared innovation. GARCH(1,1) adds a constant: ht = omega + alpha × error(t−1)² + beta × h(t−1). Under standard conditions with alpha + beta below one, long-run variance is omega/(1 − alpha − beta). Large alpha creates a stronger immediate response to news; large beta creates persistence. EWMA is a special limiting form with weights summing to one and no finite mean-reverting variance target in that specification. Both model volatility clustering without predicting return direction.
Worked example: A GARCH model has omega 0.000002, alpha 0.05, and beta 0.90. Its long-run variance is 0.00004. A large squared return raises next-period variance, which subsequently tends back toward that level if shocks moderate.
Common mistake: Applying the variance recursion directly to standard deviations mixes quantities and produces a different model from the specified EWMA or GARCH process.
Official syllabus reference: [2]
28. Bootstrap resampling and dependence
The bootstrap approximates an estimator's sampling distribution by repeatedly resampling observed data and recalculating the statistic. An ordinary nonparametric bootstrap draws observations with replacement, preserving the empirical distribution while allowing repeated selections. It is useful when analytic standard errors are difficult. Independent resampling breaks time dependence, so block methods can be more appropriate for serially related observations. Bootstrap results remain limited by the original sample: unobserved crisis events are not created automatically. Tail statistics can be particularly unstable when very few extreme observations are available to resample.
Worked example: An analyst estimates uncertainty around a median trading cost using 500 observed trades. Each bootstrap sample contains 500 draws with replacement, and its median is recorded. The resulting distribution approximates variability in the median under the empirical sampling model.
Common mistake: Assuming resampling creates new economic regimes overlooks that a nonparametric bootstrap rearranges information already present in the observed dataset.
Official syllabus reference: [2]
29. Monte Carlo and variance reduction
Monte Carlo estimates an expectation by averaging simulated outcomes from a specified model. With independent finite-variance draws, simulation standard error typically decreases at the inverse square root of the number of draws. Antithetic variates pair negatively related simulations, while control variates use a correlated quantity with known expectation to reduce estimator variance. These techniques improve computational precision without correcting an unsuitable model. Importance sampling changes the sampling distribution and reweights outcomes, which can help rare-event estimation if weights are handled correctly. Reproducible random seeds assist comparison but do not make estimates exact.
Worked example: A simulation price has estimated standard error of USD 0.08 using 10,000 draws. Using 40,000 comparable independent draws should reduce that error to about USD 0.04. A useful control variate may achieve similar precision with fewer draws.
Common mistake: Believing four times as many simulations quarters the standard error ignores the square-root convergence rate of ordinary Monte Carlo averaging.
Official syllabus reference: [2]
30. Overfitting, regularization, and validation
Overfitting occurs when a model captures sample-specific noise that does not generalize. Increasing flexibility can reduce training error while increasing out-of-sample error. Regularization constrains model complexity, trading some bias for lower estimation variance. Training data fit parameters, validation data guide model selection, and a held-out test set evaluates the final choice. For time-dependent financial data, splits should respect information availability to prevent leakage from the future. Classification accuracy alone can be misleading for rare defaults, so evaluation should reflect errors, calibration, and the intended economic decision.
Worked example: A default classifier labels every borrower as surviving and achieves 99% accuracy in a sample with 1% defaults. It still identifies no defaults, so the analyst examines recall, precision, probability calibration, and decision costs before judging usefulness.
Common mistake: Selecting the model with the lowest training error rewards memorization and can produce weaker forecasts on genuinely unseen financial observations.
Official syllabus reference: [2]
Financial Markets and Products
31. Forward and futures cash-flow timing
A forward typically fixes a future transaction price with settlement governed by its contract, while exchange-traded futures generally realize gains and losses through daily settlement. Standardization and clearing distinguish many futures arrangements from bilateral forwards. Even with the same underlying and maturity, forward and futures prices can differ when interest rates are stochastic because settlement cash flows are reinvested or funded at varying rates. Contract value also differs from delivery price: a newly initiated fair forward has approximately zero value, but an existing forward gains or loses value as market conditions change.
Worked example: A buyer enters a forward at USD 100. Later, an otherwise matching new forward trades at USD 108. The existing buyer's contract has positive value because it preserves the lower purchase price, subject to discounting and counterparty considerations.
Common mistake: Calling the agreed forward delivery price the contract's current market value confuses the future transaction amount with today's value of the agreement.
Official syllabus reference: [2]
32. Clearing, variation margin, and initial margin
A central counterparty interposes itself between clearing participants and manages their exposures through margin, settlement, and other resources. Variation margin reflects changes in marked exposure, while initial margin provides protection against potential adverse moves over a closeout period. Clearing transforms counterparty relationships but does not eliminate market risk, liquidity demands, or every form of counterparty risk. A participant may need cash for settlement before receiving offsetting proceeds elsewhere. Margin rules depend on the relevant arrangement, so numerical maintenance and restoration requirements should be supplied rather than assumed universally.
Worked example: A futures position loses USD 12,000 during the day. Its holder must meet the resulting settlement obligation even if an offsetting physical asset gains value on paper, because that physical gain may not immediately provide cash.
Common mistake: Treating margin as the maximum possible loss overlooks additional settlement demands and losses that can exceed the amount initially posted.
Official syllabus reference: [2]
33. Cost-of-carry forward pricing
For a tradable investment asset with continuous financing rate r and continuous income yield q, the basic forward price is S × exp[(r − q)T]. Financing increases carrying cost, while income earned from ownership reduces it. Known cash income can instead be subtracted from spot at present value before financing the remaining amount. The relationship follows from comparing forward ownership with a financed cash position. Transaction costs, short-sale constraints, funding differences, and counterparty terms can create pricing bands or adjustments. Delivery price and the value of an existing forward must remain distinct.
Worked example: An index is worth 200, financing costs 5% continuously, and dividends yield 2%. A one-year fair forward is 200 × exp(0.03), approximately 206.09, because ownership earns dividends that offset part of the financing cost.
Common mistake: Adding dividends to financing cost reverses their economic effect because the owner of the underlying receives that income during the holding period.
Official syllabus reference: [2]
34. Covered interest parity and FX quotes
When an exchange rate is quoted as domestic currency per unit of foreign currency, covered interest parity gives F = S × exp[(rd − rf)T] under continuous compounding and idealized funding assumptions. Buying foreign currency spot and investing it must match the domestic-funded forward alternative. Quote orientation matters: reversing the currency pair reverses which rate belongs in the numerator relationship. A forward premium is an interest differential implication, not a direct forecast of currency appreciation. Actual funding, collateral, and cross-currency basis conditions can modify the simple textbook relationship.
Worked example: Suppose spot is 1.20 domestic units per foreign unit, domestic interest is 4%, and foreign interest is 1%. The one-year forward is approximately 1.2365 domestic units per foreign unit under the stated parity assumptions.
Common mistake: Using the foreign-minus-domestic rate difference with a domestic-per-foreign quote reverses the carry relationship and produces the reciprocal convention's adjustment.
Official syllabus reference: [2]
35. Minimum-variance hedge ratios
A futures hedge minimizes variance when its exposure ratio is correlation between spot and futures changes multiplied by spot-change volatility divided by futures-change volatility. Contract count then scales that ratio by exposure quantity divided by contract quantity, using compatible units. A short hedge protects an asset to be sold; a long hedge protects a future purchase. A one-for-one physical hedge need not minimize variance when price changes differ. Under the basic single-regressor model, the optimized variance reduction is squared correlation. Estimated ratios can change with horizon, sample, and market conditions.
Worked example: A buyer needs 80,000 units. Correlation is 0.75, spot-change volatility is 4, futures-change volatility is 3, and contracts cover 2,000 units. The hedge ratio is one, so the buyer goes long 40 contracts.
Common mistake: Using correlation alone as the hedge ratio ignores the relative volatility of the two price changes and can materially mis-size the position.
Official syllabus reference: [2]
36. Clean prices, dirty prices, and accrued interest
A bond's dirty price is the cash settlement price including accrued interest; its clean price excludes that accrued amount. Accrued interest allocates coupon income between payment dates according to the instrument's specified convention. Day-count rules determine the fraction of the coupon period that has elapsed and can differ across instruments. Quoted price changes should be interpreted consistently with the pricing convention. Yield calculations use actual settlement cash flows and dates rather than treating accrued interest as an unrelated fee. Special settlement conventions require explicit treatment when relevant.
Worked example: A bond with USD 1,000 face value has a clean price of 101% of face. Its semiannual coupon is USD 30, and one third of the coupon period has elapsed. Accrued interest is USD 10 and settlement is USD 1,020.
Common mistake: Adding the entire next coupon to the clean price overstates settlement because only the accrued portion belongs to the seller at that date.
Official syllabus reference: [2]
37. Repo financing and collateral haircuts
A repurchase agreement economically combines secured cash financing with a sale and agreed repurchase of collateral. The repo rate determines the financing charge over the term using the contract's accrual convention. A haircut makes cash advanced lower than collateral market value, providing protection against adverse price changes and liquidation costs. Haircuts do not eliminate counterparty or market exposure. Changes in collateral value or required haircuts can create liquidity demands. Specific securities may trade at special repo rates when demand to obtain that collateral differs from general cash-financing conditions.
Worked example: A dealer finances securities worth USD 20 million with a 5% haircut, receiving USD 19 million cash. If the required haircut rises while collateral value remains unchanged, the dealer must supply additional resources or reduce financing.
Common mistake: Applying the haircut to the interest payment rather than collateral value misunderstands how the lender determines the initial cash advance.
Official syllabus reference: [2]
38. Forward rate agreements and settlement discounting
A forward rate agreement fixes interest on a notional borrowing or lending period. For a party receiving floating and paying fixed, the end-of-period interest difference is notional times accrual fraction times the reference-rate-minus-fixed-rate spread. If settlement occurs at the beginning of the underlying period, that difference is discounted according to the specified convention. The notional itself is generally not exchanged. Sign depends on which rate the party receives, and timing determines whether discounting is required. The contract's reference rate, dates, and day-count basis must be supplied for an exact calculation.
Worked example: A receive-floating agreement has USD 10 million notional, a quarter-year accrual, fixed rate 4%, and reference rate 5%. The period-end difference is USD 25,000. Under simple beginning settlement at 5%, the receipt is 25,000 divided by 1.0125.
Common mistake: Paying the undiscounted interest difference at the period's start overlooks that the recipient receives funds earlier than the underlying interest date.
Official syllabus reference: [2]
39. Swap cash flows and economic exposure
A plain interest rate swap exchanges fixed and floating interest amounts on a notional principal. In a single-currency arrangement, principal usually scales payments without being exchanged. Receiving floating and paying fixed can convert floating-rate borrowing into approximately fixed-rate borrowing when reset dates, notionals, and reference rates align. The current net coupon differs from the swap's market value, which reflects all remaining projected and discounted payments. Basis mismatch can remain when the debt and swap reference different rates. Credit, collateral, and discounting arrangements also affect valuation beyond the basic cash-flow identity.
Worked example: A borrower pays a reference rate plus 1% on debt. It enters a swap receiving that same reference rate and paying 3%. Matching terms leave an approximate combined interest cost of 4%, before fees and other adjustments.
Common mistake: Assuming every pay-fixed swap perfectly fixes borrowing costs ignores mismatches in reference rates, payment dates, principal schedules, and financing spreads.
Official syllabus reference: [2]
40. Bond futures delivery and cheapest-to-deliver
Some bond futures allow the short position to choose among eligible deliverable bonds. Conversion factors adjust quoted futures prices for differences between those bonds, but do not make delivery economics identical. The cheapest-to-deliver bond is determined by comparing eligible bonds' cash prices with delivery proceeds while accounting for relevant carry and timing. A common simplified net basis compares clean cash price with futures price times conversion factor. Accrued interest enters invoice calculations under the contract. Delivery options can affect futures valuation and hedge performance, especially when the preferred deliverable changes.
Worked example: Two eligible bonds have clean prices of 104 and 108, with conversion factors 1.02 and 1.05. At a futures price of 100, simplified net bases are two and three price points before carry, favoring the first bond.
Common mistake: Selecting the lowest cash-price bond without conversion factors and carry ignores the delivery proceeds that determine which bond is economically cheapest.
Official syllabus reference: [2]
41. Option payoff, profit, and exercise
A call gives its holder the right to buy at the strike, while a put gives the right to sell. At expiry, call payoff is max(S − K,0) and put payoff is max(K − S,0). Profit also accounts for the premium and, when specified, its financing cost. European options exercise only at maturity; American options permit earlier exercise under their terms. An option can have a positive payoff yet produce a negative investment profit. Before expiry, time value and exercise features matter, so intrinsic value alone generally does not determine market price.
Worked example: A call costs USD 6 and has strike USD 50. If the stock ends at USD 54, payoff is USD 4 but the buyer loses USD 2 before financing and costs. The option finishes in the money despite an investment loss.
Common mistake: Calling every in-the-money option profitable ignores the premium paid and any financing or transaction costs associated with establishing the position.
Official syllabus reference: [2]
42. Put-call parity and synthetic positions
For European options on a non-dividend-paying stock with matching strike and maturity, call plus discounted strike equals put plus stock. The portfolios have identical expiry values, creating a no-arbitrage relationship. Rearranging constructs synthetic exposures, such as long call plus short put replicating a forward-style payoff. Income and exercise conventions require appropriate modifications. Parity compares portfolios, not isolated options, and assumes the financing and trading operations are feasible. A seeming violation may disappear after accounting for bid-ask spreads, borrowing costs, dividends, or different contract specifications.
Worked example: A stock costs USD 60, the discounted strike is USD 58, and a matching put costs USD 3. Parity implies a USD 5 call. Buying the stock and put therefore costs the same as the call plus strike funding.
Common mistake: Applying basic European parity unchanged to mismatched maturities or dividend assumptions compares portfolios whose future cash flows are no longer identical.
Official syllabus reference: [2]
43. Vertical spreads and bounded payoffs
A bull call spread buys a lower-strike call and sells a higher-strike call on the same underlying and expiry. Its maximum expiry payoff equals the strike difference, while its initial net premium determines maximum loss and reduces maximum profit. Selling the higher call finances part of the purchase but gives up gains above that strike. A bear put spread reverses the directional objective using puts. Strategy analysis should combine each leg's piecewise payoff before subtracting premiums. Exercise style and settlement terms can create additional operational considerations before expiry.
Worked example: An investor buys a strike-100 call for USD 8 and sells a strike-110 call for USD 3. The net cost is USD 5. Maximum expiry payoff is USD 10, so maximum profit and maximum loss are each USD 5.
Common mistake: Describing a bull call spread as having unlimited upside ignores the short higher-strike call that caps the combined expiry payoff.
Official syllabus reference: [2]
44. Commodity storage and convenience yield
Commodity forward pricing reflects financing, storage costs, and benefits of physically holding inventory. Convenience yield represents those noncash ownership benefits, such as avoiding production interruption when supplies are scarce. In a simplified continuous-yield model, forward price is S × exp[(r + storage yield − convenience yield)T]. High convenience yield can push futures below spot despite positive financing costs. Physical constraints and limited short selling can weaken simple cash-and-carry bounds. Contango and backwardation describe price-curve shapes; neither alone establishes an unbiased forecast of future spot prices.
Worked example: A manufacturer values immediate access to scarce metal because an inventory shortage would stop production. That convenience benefit can outweigh financing and storage costs, making near-term futures cheaper than spot even when interest rates are positive.
Common mistake: Interpreting backwardation as proof that market participants expect spot prices to fall ignores inventory benefits and the risk premiums embedded in futures prices.
Official syllabus reference: [2]
45. Securitization waterfalls and tranche losses
Securitization pools asset cash flows and allocates them across claims according to a contractual waterfall. Subordination directs initial credit losses to junior claims before senior claims, redistributing risk without eliminating pool losses. A tranche's attachment and detachment points identify the pool-loss interval it absorbs. Dependence among underlying defaults strongly influences the distribution of losses across tranches. Prepayment, timing, servicing, and structural triggers can also affect outcomes. Seniority protects against specified loss layers but does not make a claim immune to severe common shocks or uncertainty about underlying asset quality.
Worked example: A tranche absorbs pool losses between 5% and 10%. If cumulative pool loss reaches 7%, the tranche absorbs two percentage points of pool principal, equivalent to 40% of its own five-percentage-point thickness, under a simple loss allocation.
Common mistake: Treating a senior tranche as free of credit risk overlooks severe pool losses and the dependence assumptions supporting its apparent protection.
Official syllabus reference: [2]
Valuation and Risk Models
46. Discount factors and spot-curve bootstrapping
A discount factor converts a dated cash flow into present value. A bond's value is the sum of each contractual cash flow multiplied by the discount factor for its payment date. Bootstrapping extracts successive discount factors from traded instruments by first using prices that determine earlier dates, then solving for later dates. A single yield to maturity summarizes one bond's price but does not generally equal every relevant spot rate. Curve construction requires consistent cash-flow timing, quotation, interpolation, and accrual conventions. Different assumptions can produce different intermediate discount factors even when quoted instruments are repriced.
Worked example: A one-year discount factor is 0.97. A two-year bond costs 101 and pays five after one year plus 105 after two years. Solving 101 = 5 × 0.97 + 105 × D2 gives D2 approximately 0.9157.
Common mistake: Discounting every cash flow at one bond's yield to maturity treats its internal rate as a universal spot curve for other instruments.
Official syllabus reference: [2]
47. Modified duration and DV01
Modified duration approximates proportional price change for a small yield change: delta P/P is approximately minus duration times delta yield. DV01 converts that sensitivity into currency value for a one-basis-point move, commonly reported as positive magnitude for a standard long bond. Approximate DV01 equals modified duration times price times 0.0001. Macaulay duration is a cash-flow-weighted time measure and differs from modified duration through the yield-compounding adjustment. Sign conventions should be stated. Duration assumes the relevant cash-flow and yield relationship remains locally stable, limiting its usefulness for large moves or changing embedded-option behavior.
Worked example: A portfolio worth USD 5 million has modified duration of six. Its DV01 magnitude is USD 3,000. A parallel rise of three basis points therefore implies an approximate USD 9,000 loss before convexity and other effects.
Common mistake: Multiplying by 0.01 for a one-basis-point move overstates the sensitivity one hundredfold because a basis point is 0.0001.
Official syllabus reference: [2]
48. Convexity and nonlinear bond sensitivity
Convexity measures curvature in the relationship between bond price and yield. A second-order approximation adds one half times convexity times squared yield change to the duration-based proportional price change. For a typical option-free bond with positive convexity, this term increases estimated price relative to the straight-line approximation for both upward and downward yield moves. Embedded options can alter or reverse convexity over relevant rate ranges. Convexity conventions must match the yield and time units used. The approximation remains local and does not replace full repricing for large or structurally complex changes.
Worked example: A bond has duration five and convexity 40. For a 1% yield increase, proportional change is approximately −5% + 0.5 × 40 × 0.01² = −4.8%. Convexity offsets part of the duration-only loss.
Common mistake: Changing the sign of the convexity term when yields fall ignores that the second-order adjustment contains the squared yield change.
Official syllabus reference: [2]
49. Key-rate duration and curve-shape risk
Key-rate duration measures sensitivity to a localized movement at a selected maturity point, with a specified interpolation rule for the surrounding curve. A collection of key-rate sensitivities describes exposure to twists and other nonparallel changes that aggregate duration can hide. Two portfolios can have the same overall duration yet very different maturity concentrations. Hedge instruments should therefore be selected to address the relevant curve locations. Key-rate results depend on curve construction and shock definition; sensitivities from different systems are not automatically comparable without consistent bump sizes and interpolation conventions.
Worked example: A portfolio is long short-maturity bonds and short long-maturity bonds, with near-zero aggregate duration. It can still lose when short yields rise and long yields fall, because opposite key-rate exposures reinforce under that curve twist.
Common mistake: Calling a zero-duration portfolio interest-rate neutral overlooks curve-shape changes that move maturity segments differently and can generate substantial residual losses.
Official syllabus reference: [2]
50. Binomial replication and risk-neutral pricing
A binomial model values a derivative by matching its up-state and down-state payoffs with stock and risk-free financing. The equivalent risk-neutral probability is the risk-free growth factor minus the down factor, divided by the up-minus-down difference. Discounting expected payoff under these pricing probabilities gives the no-arbitrage value. The risk-free growth factor must lie between the stock factors in the basic model. Actual expected stock returns do not enter this replication price. For American options, backward induction compares continuation value with immediate exercise at each eligible node.
Worked example: A stock moves from 50 to either 60 or 45, and the gross risk-free factor is 1.02. The risk-neutral up probability is 0.40. A strike-50 call pays ten or zero, so its value is 4/1.02, approximately 3.92.
Common mistake: Using an analyst's estimated real-world up probability in the replication formula mixes forecasting probabilities with no-arbitrage pricing weights.
Official syllabus reference: [2]
51. Black-Scholes-Merton assumptions and interpretation
The basic Black-Scholes-Merton framework prices European options using a continuous diffusion with constant volatility and interest rate, frictionless trading, and continuous rebalancing assumptions. In its non-dividend stock form, the call value is S times N(d1) minus discounted strike times N(d2). Risk-neutral valuation removes the stock's actual expected return from the pricing formula. Market-implied volatility is the volatility input that matches an observed option price. Different implied volatilities across strikes or maturities reveal departures from the constant-volatility ideal, rather than a single universally applicable volatility forecast.
Worked example: Two same-expiry options on one stock require implied volatilities of 20% and 28% at different strikes. The discrepancy forms part of an implied-volatility smile or skew, signaling that one constant-volatility input does not fit both prices.
Common mistake: Treating implied volatility as a guaranteed future realized volatility confuses a price-inverting model input with an objectively certain forecast.
Official syllabus reference: [2]
52. Delta, gamma, vega, and theta
Option Greeks describe local sensitivity to specific inputs. Delta is the first derivative with respect to underlying price; gamma measures how delta changes with that price. Vega measures sensitivity to volatility, and theta measures sensitivity to the passage of time under a stated convention. Rho measures interest-rate sensitivity. A local approximation can combine these effects, but units matter: vega may be quoted per one percentage-point volatility change rather than per unit decimal volatility. Greeks change with market conditions, so one snapshot does not characterize the position's behavior across every scenario.
Worked example: An option position has vega USD 12,000 per volatility percentage point. If implied volatility rises from 20% to 22%, its first-order vega gain is approximately USD 24,000, holding other inputs fixed and ignoring higher-order effects.
Common mistake: Multiplying a per-percentage-point vega by 0.02 for a two-point volatility increase mixes quotation conventions and understates the estimated change.
Official syllabus reference: [2]
53. Delta-gamma hedging and rebalancing
A delta hedge offsets first-order sensitivity to the underlying, typically using the underlying asset or futures. It does not remove gamma, so a price move changes the hedge's delta. Another option can adjust gamma because the underlying itself has zero gamma in the basic framework. After choosing an option quantity to offset gamma, the remaining delta is hedged separately. Vega, jumps, transaction costs, and discrete rebalancing can still create losses. A locally neutral portfolio is therefore not globally riskless, particularly when market changes are large or several inputs move together.
Worked example: A position has gamma of 600 and delta of 900. An available option has gamma three and delta 0.4 per unit. Shorting 200 options offsets gamma and leaves delta 820, which is offset by shorting 820 underlying units.
Common mistake: Assuming a delta-neutral position remains neutral after a large price move ignores gamma and the need to reassess the hedge.
Official syllabus reference: [2]
54. Value at risk as a loss quantile
Value at risk is a quantile of a specified loss distribution over a specified horizon. A 99% VaR identifies a threshold exceeded with approximately 1% probability under the model, subject to discrete-distribution conventions. For linear exposure and normal returns, a volatility-based formula can be used with an explicit mean and loss convention. VaR does not describe the magnitude of losses beyond its threshold. Scaling by the square root of time relies on restrictive assumptions about return dependence and variance stability. Nonlinear positions and changing volatility can invalidate simple aggregation or scaling.
Worked example: A model reports one-day 99% VaR of USD 400,000. Under its assumptions, losses exceed that threshold on about 1% of comparable days. The figure does not say that losses beyond USD 400,000 are impossible or small.
Common mistake: Describing VaR as the maximum possible loss replaces a probabilistic threshold with a bound that the model never established.
Official syllabus reference: [2]
55. Expected shortfall and tail averaging
Expected shortfall averages losses across the specified worst tail probability. Unlike a single quantile, it responds to how severe losses become beyond the VaR threshold. For continuous distributions it can be expressed as conditional average loss beyond VaR. With discrete distributions, averaging exactly the worst tail may require only part of the probability mass at the boundary. This distinction avoids accidentally averaging a larger or smaller tail. Expected shortfall supports diversification under standard coherent-risk definitions, but its estimation can be difficult because few observations populate the most severe tail.
Worked example: Losses of USD 5 million occur with 1% probability and losses of USD 2 million with 4%. The worst 2% contains all the first group and one percentage point of the second, giving expected shortfall of USD 3.5 million.
Common mistake: Averaging every observation equal to or above a discrete VaR can include too much probability mass and misstate the specified tail average.
Official syllabus reference: [2]
56. Historical simulation versus model-based simulation
Historical simulation applies observed risk-factor changes to current positions and ranks resulting losses. It preserves relationships present in the sampled scenarios but depends heavily on the sample and its relevance. Model-based Monte Carlo generates changes from a specified joint distribution, allowing more scenarios and explicit dependence assumptions. Full repricing captures nonlinear payoffs better than a simple linear sensitivity approximation, though it can be computationally expensive. Neither method automatically includes plausible events absent from its data or model. Scenario weighting, volatility adjustment, and horizon choices materially affect the resulting loss distribution.
Worked example: A portfolio now contains a large option position. Reusing historical underlying changes with full option repricing captures changing payoff curvature more faithfully than multiplying every historical move by today's delta, although volatility shocks must also be modeled.
Common mistake: Assuming historical simulation requires no modeling overlooks choices about data windows, factor mapping, repricing, missing observations, and the relevance of past conditions.
Official syllabus reference: [2]
57. VaR backtesting: coverage and independence
Backtesting compares realized losses with forecasts made before those losses occurred. A VaR exception occurs when realized loss exceeds the forecast threshold. Unconditional coverage examines whether the exception rate is consistent with the stated confidence level; independence examines whether exceptions cluster in time. A model can have the right total count while missing changing volatility and producing clustered failures. The profit-and-loss definition should match the model's scope and horizon. Tests have sampling uncertainty, so passing one backtest does not establish that assumptions, tail severity, or future performance are adequate.
Worked example: A 99% VaR model records ten exceptions over 1,000 days, but eight occur during one short crisis episode. The total count appears consistent with nominal coverage, while clustering suggests the model adjusts too slowly to changing risk.
Common mistake: Checking only the number of exceptions overlooks whether failures cluster precisely when timely risk forecasts matter most to the institution.
Official syllabus reference: [2]
58. Stress testing and reverse stress testing
Stress testing evaluates losses under specified severe conditions, using historical episodes, hypothetical scenarios, or targeted factor shocks. Scenarios should connect interacting variables coherently rather than combine arbitrary extremes without an economic narrative. Reverse stress testing begins with a defined failure condition and identifies circumstances capable of causing it. Neither method requires assigning a precise probability to every scenario. Stress analysis complements distribution-based measures by exploring model weaknesses, concentration, liquidity demands, and nonlinear effects. Results depend on portfolio response assumptions, including whether proposed hedges or asset sales remain feasible in the stressed market.
Worked example: A reverse stress exercise defines failure as exhausting a USD 30 million liquidity buffer. Analysts identify combinations of collateral calls, funding withdrawals, and asset-sale discounts that cross that boundary, then examine which combinations are economically plausible.
Common mistake: Treating an unassigned scenario probability as a reason to ignore stress results overlooks their purpose in revealing vulnerabilities outside routine statistical forecasts.
Official syllabus reference: [2]
59. PD, LGD, EAD, and credit-loss dependence
Probability of default measures the chance of default over a defined horizon. Loss given default is the fraction of exposure lost conditional on default, and exposure at default includes amounts expected to be outstanding at that time. With fixed scenario inputs, expected loss is PD times LGD times EAD. In richer models, exposure and severity can depend on default conditions, so multiplying unconditional averages may be inadequate. Unexpected loss concerns dispersion around expected loss and depends on concentration and default dependence. Recovery timing and discounting can also affect the measured severity.
Worked example: A facility has USD 4 million drawn and USD 2 million undrawn. Assuming half the undrawn amount is used before default gives EAD of USD 5 million. With PD 2% and LGD 30%, expected loss is USD 30,000.
Common mistake: Using only today's drawn balance understates exposure when borrowers are expected to draw additional committed funds before or during financial distress.
Official syllabus reference: [2]
60. Operational loss frequency and severity
Operational-loss modeling can separate the number of events from the size of each event. A frequency distribution describes how many losses occur, while a severity distribution describes loss amounts conditional on an event. Aggregate annual loss sums the severities across the random event count. Under independence and identical severities, expected aggregate loss equals expected frequency times expected severity. Tail risk depends on more than those averages, especially with heavy-tailed severities or shared causes. Sparse data, reporting thresholds, changing controls, and external events complicate estimation and require explicit assumptions about data relevance.
Worked example: A process averages four incidents per year, each with mean loss USD 25,000. Under the stated independent frequency-severity model, expected annual loss is USD 100,000. A rare major incident can still dominate the upper tail despite that moderate average.
Common mistake: Multiplying average frequency and severity to estimate a high loss percentile confuses the mean aggregate loss with the distribution's extreme outcomes.
Official syllabus reference: [2]
Official syllabus sources
- FRM® Exam Information, Steps to Earn Certification | GARP
- FRM® Study Material, Guide, Books, Practice Exams | GARP