How do value at risk and expected shortfall differ?

Value at risk identifies a loss threshold at a stated horizon and confidence level. Expected shortfall estimates the average loss beyond that threshold under a specified method. For hypothetical losses of 8, 10, 12, 15, and 25 million in the selected tail, expected shortfall is 14 million. The 25 million observation affects that average, even if it leaves the value-at-risk threshold unchanged. Both measures depend on the loss distribution and its assumptions.

Why can a solvent firm still face a liquidity crisis?

Solvency concerns whether assets exceed liabilities in value. Funding liquidity concerns whether cash arrives when payments are due. A firm may hold valuable long-term loans yet owe 100 million tomorrow while only 40 million of cash is available. It must raise the 60 million gap by selling assets, borrowing, or using a committed source of funds. A rushed sale can reduce asset values and turn a liquidity problem into a solvency problem.

How should a risk manager use a model result?

First identify the decision and the risk the model is meant to measure. Then check the inputs, horizon, assumptions, and conditions in which the model has performed poorly. Compare the result with limits, stress scenarios, and relevant observed losses. If a model reports a small market loss while a plausible rate shock produces a large loss, investigate the difference before increasing the position. A model result informs judgment; it does not remove the need to examine exposures and controls.

Financial Risk Manager (FRM) Study Guide

Start with concept 1 Choose a topic

Risk management starts with a practical question: what could change, how much could it cost, and what action would reduce the exposure? The FRM subjects build from that question. Part I develops the statistical tools, financial products, and valuation methods used to describe risk. Part II applies them to market, credit, operational, liquidity, and investment decisions, then considers risks that can spread across firms and markets. As you work through this guide, keep the assumptions beside every number. A loss estimate depends on its horizon, confidence level, data, and model; a hedge depends on what actually drives the exposure. The examples use stated, hypothetical figures so you can check both the arithmetic and the decision it supports.

Foundations of Risk Management

1. Identify the source of risk

Classify a loss by the event that produces it, then choose a control that addresses that event. Market risk arises when prices, rates, or exchange rates move against a position. Credit risk arises when an obligor fails to meet a payment obligation or becomes less likely to do so. Liquidity risk appears when cash cannot be raised on time or an asset can be sold only at a steep discount. Operational risk arises from failed processes, people, systems, or external events. One incident can involve several types: a system outage may delay settlement, creating a cash shortage. Trace the sequence instead of forcing the whole incident into one label. The FRM program includes foundations of risk management among its Part I topics.

Worked example: Suppose a bond loses value when rates rise. The trigger is a market move, so rate sensitivity limits or a rate hedge address it. If its issuer misses a coupon, the trigger is credit failure, calling for issuer exposure limits and credit review. If a settlement system fails while transferring the bond, the trigger is operational, calling for system recovery and reconciliation controls.

Watch the distinction: A bond loss is not automatically credit risk. Identify what changed first: the market price, the issuer's ability to pay, or the process used to hold or transfer it.

Reference: [1]

2. Turn appetite into limits

Risk appetite states how much uncertainty an institution is willing to accept in pursuit of its objectives. A useful limit turns that broad statement into a measurable boundary, such as maximum exposure to one sector, counterparty, or market factor. Define the exposure measure before testing a trade: gross and net positions, current value and potential future exposure, and currency conversion can produce different answers. Then compare the proposed post-trade position with the approved limit, not just the existing position. Monitoring needs a named owner, timely data, and an escalation route for a proposed or actual breach. A warning level below the limit can create time to act. A limit is a decision constraint; it does not make positions inside it safe under every scenario.

Worked example: A hypothetical sector limit is $50 million, and current exposure is $44 million. Capacity is $50 million − $44 million = $6 million. An $8 million purchase would raise exposure to $52 million, exceeding the limit by $2 million. The desk can reduce the purchase to at most $6 million or seek an authorized exception before executing it.

Watch the distinction: Comparing an $8 million purchase with a $50 million limit ignores the $44 million already held. Limits apply to the measured total after the proposed trade.

Reference: [1]

3. Assign risk decisions clearly

A control structure works when each party knows what it owns. The business unit creates or accepts an exposure and must operate within approved authority, record positions accurately, and correct breaches. An independent risk function sets or monitors risk measures, challenges assumptions and decisions, and escalates concerns without taking over the business decision. Internal audit provides later assurance by testing whether the governance and controls were designed and operated effectively. These roles should communicate, but independence matters: a unit should not be the only judge of its own compliance. For an incident, ask three questions in order: who took the risk, who should have challenged it, and who verifies that the control system worked? Accountability remains with the risk owner even when another function detects the problem.

Worked example: A trader opens a position without approval. The trading business owns the exposure and must report, reduce, or obtain proper authorization for it. Independent risk checks the exposure against limits and challenges any justification for keeping it. Internal audit later examines approvals, monitoring records, and escalation evidence to determine whether the controls worked. Detection by risk does not transfer ownership away from the business.

Watch the distinction: Internal audit does not approve a trader's position after the fact. Its assurance role is to assess the control system; the business and authorized decision makers resolve the exposure.

Reference: [1]

4. Recognize diversification limits

Combining positions can reduce volatility when their returns do not move together. For two assets, portfolio variance equals each weighted variance plus twice the product of their weights, volatilities, and correlation. The correlation term determines how much the movements reinforce one another. With positive correlation below one, diversification lowers volatility relative to a simple weighted average; at correlation one, there is no such reduction for two positive positions. Negative correlation can provide a stronger offset. This calculation describes variability under the assumed volatilities and correlation, not a guaranteed loss limit. Dependence can change in stress, and ordinary correlation may miss simultaneous extreme losses. Before relying on diversification, examine shared economic drivers and whether the positions could become hard to sell at the same time.

Worked example: Two hypothetical positions each have 10% volatility and a 50% portfolio weight. At zero correlation, variance is 0.5² × 10² + 0.5² × 10² = 50 percentage-points squared, so volatility is √50 = 7.07%. At correlation one, add 2 × 0.5 × 0.5 × 10 × 10 = 50; volatility becomes √100 = 10%.

Watch the distinction: Owning two securities does not by itself halve risk. Their weights, individual volatility, and dependence determine the result, and that dependence may increase during a market shock.

Reference: [1]

5. Trace systemic spillovers

Systemic risk concerns the transmission of distress beyond the firm where a problem begins. A counterparty loss can deprive another firm of expected cash. To meet its own obligations, that firm may sell assets quickly; those sales can push market prices down. Lower prices then reduce other investors' wealth and the value of collateral supporting loans, prompting margin calls or tighter funding. The diagnostic task is to map obligations, timing, and shared assets. Distinguish the original loss from the channel that spreads it: missed payment, funding withdrawal, asset sale, or collateral revaluation. Feedback makes the final effect larger than the first unpaid amount because affected firms may respond at once. Concentrated counterparties and common holdings make this chain more likely to reach otherwise sound firms.

Worked example: A dealer misses a hypothetical $20 million payment to a lender. The lender now has a $20 million expected-cash shortfall and sells bonds to fund payments due today. If that sale depresses bond prices, other firms' collateral is marked lower and may trigger calls for more cash. The forced sale and collateral revaluation spread the dealer's initial failure beyond its direct lender.

Watch the distinction: The missed $20 million is only the starting exposure. Counting that payment alone misses the funding actions and price changes that can transmit distress to other firms.

Reference: [1]

6. Compare return with risk capital

Risk-adjusted return on capital, or RAROC, compares a business activity's earnings with the capital assigned to absorb its risk. A common form is risk-adjusted income divided by economic capital. State which income adjustments and capital method you use: expected credit losses, operating costs, taxes, and diversification can change the comparison. A higher percentage means more measured return per unit of assigned capital, provided both activities use consistent definitions and horizons. It does not automatically mean the activity is preferable. An unusually high estimate may reflect understated tail risk, a short observation period, or benefits that are hard to measure. Use RAROC alongside concentration, liquidity, and strategic constraints. When checking arithmetic, divide currency by currency so the result is a percentage, then compare it with the institution's required return.

Worked example: Desk A earns a hypothetical $12 million after expected losses and uses $100 million of risk capital: RAROC = 12/100 = 12%. Desk B earns $9 million on $50 million: 9/50 = 18%. B earns more per unit of measured capital, although A contributes $3 million more absolute income. A decision still depends on consistent capital estimates and other constraints.

Watch the distinction: Choosing Desk A solely because $12 million exceeds $9 million ignores capital use. Equally, choosing B solely from RAROC ignores whether its capital estimate captures its risks.

Reference: [1]

Quantitative Analysis

7. Update a probability with evidence

A signal's detection rate is the chance of a positive flag when the event is present. The probability a flagged case actually contains the event is different because flags also occur among cases without the event. Bayes' rule combines three inputs: prior event prevalence, detection rate, and false-positive rate. Multiply prevalence by detection rate to obtain true flags as a share of all cases. Multiply the non-event share by the false-positive rate to obtain false flags. Divide true flags by all flags. When the event is rare, even a useful signal can have many false alarms because the non-event population is large. Keep the population and time horizon consistent. For further practice on this Part I tool, use the FRM Part I study guide.

Worked example: In a hypothetical group of 10,000 borrowers, 2%, or 200, default. An 80% detection rate flags 160 of them. Of the 9,800 who do not default, a 10% false-positive rate flags 980. Thus 1,140 are flagged, and P(default | flag) = 160/1,140 ≈ 14.0%. The flag raises concern above the 2% prior but does not imply 80% default probability.

Watch the distinction: The 80% detection rate answers how often defaulters are flagged. It does not answer how often flagged borrowers default; false flags and default prevalence belong in the denominator.

Reference: [1]

8. Calculate expected loss variability

The expected value is the probability-weighted average outcome. Variance asks how far outcomes tend to sit from that mean: subtract the mean from each outcome, square the difference, multiply by its probability, and add the results. Standard deviation is the square root of variance, returning the measure to the original unit. When outcomes are returns, using percentage-point values produces variance in percentage-points squared and standard deviation in percent. Alternatively, use decimal returns throughout; the standard deviation then appears as a decimal. This is a population calculation for a stated probability distribution, not a sample variance estimate from observed data. Variability treats upward and downward deviations symmetrically, so it does not alone describe downside severity. Check that the probabilities sum to one before calculating.

Worked example: Suppose hypothetical returns are 0% and 10%, each with probability 0.5. The mean is 0.5(0%) + 0.5(10%) = 5%. Variance is 0.5(0 − 5)² + 0.5(10 − 5)² = 25 percentage-points squared. Standard deviation is √25 = 5%. Either outcome lies five percentage points from the mean.

Watch the distinction: Variance is not 25% in this example. Its unit is percentage-points squared; taking the square root gives the interpretable 5% standard deviation.

Reference: [1]

9. Build portfolio variance

For two assets, portfolio volatility depends on position weights, each asset's volatility, and the correlation between returns. Square the weights and volatilities for the standalone variance terms, then add the covariance term: 2w₁w₂σ₁σ₂ρ. The covariance term is positive when the assets tend to move together and negative when they tend to offset each other. The square root of total variance gives portfolio volatility. Use one unit consistently: either decimal returns throughout or percentage-point volatilities throughout. Weights should represent the portfolio amounts whose return is being measured; leverage or short positions can make weights negative or sum in ways that need careful interpretation. A low measured correlation can reduce estimated volatility, but a shared shock may change correlation precisely when a hedge is needed most.

Worked example: Take equal weights, hypothetical volatilities of 8% and 12%, and correlation 0.5. In percentage-point units, variance is 0.25(8²) + 0.25(12²) + 2(0.5)(0.5)(0.5)(8)(12) = 16 + 36 + 24 = 76 percentage-points squared. Volatility is √76 ≈ 8.72%, below the weighted average volatility of 10%.

Watch the distinction: Averaging 8% and 12% gives 10%, but that leaves out covariance. Portfolio volatility requires squaring exposures, adding the correlation term, and taking the square root.

Reference: [1]

10. Read a distribution's tail

A tail probability describes how often an outcome falls beyond a specified cutoff under a chosen distribution. Standardizing a return converts it to a z-score: subtract the assumed mean and divide by the assumed standard deviation. For a loss threshold, be explicit about the sign of the return and whether the question asks for one tail or both. A normal distribution supplies a specific mapping from z-scores to probabilities; it is an assumption about the shape of outcomes, not something the mean and standard deviation prove. Financial returns may have more extreme observations than that shape predicts. Use the normal result as a model-based estimate and check historical extremes or stress scenarios when tail losses drive a decision. Also match the time unit of the volatility to the time unit of the event.

Worked example: Assume a hypothetical daily return is normal with mean 0% and standard deviation 2%. A loss worse than 4% means a return below −4%, so z = (−4% − 0%)/2% = −2. The normal left-tail probability is about 2.3% for one day. This is one-sided; doubling it would describe moves beyond ±4%, a different event.

Watch the distinction: Calling a two-standard-deviation loss impossible confuses a low model probability with a bound. The 2.3% estimate also depends on the assumed normal distribution.

Reference: [1]

11. Interpret a regression slope

A fitted regression separates a predicted response from a residual. In a simple linear model, the intercept is the prediction when the input is zero, and the slope is the predicted change in the outcome for a one-unit change in that input. Units matter: if the input is a rate change in percentage points and the outcome is a bond return in percent, the slope states percentage points of return per percentage point of rate movement. Substitute the observed input to get the fitted return, then subtract fitted from actual to obtain the residual. A residual can reflect other risk factors, noise, or a model that does not fit well. A slope estimated from past data describes association under that sample and specification; it should not be treated as an exact causal response or extrapolated far beyond observed changes.

Worked example: Suppose a hypothetical fitted bond return is 1% − 2 × Δr, with Δr measured in percentage points. For a 0.5-point rate rise, predicted return is 1% − 2(0.5)% = 0%. If actual return is −0.3%, the residual is −0.3% − 0% = −0.3%. The rate input explains the fitted result, while the residual remains unexplained by this model.

Watch the distinction: Substituting 0.005 when the model expects percentage points would understate the rate input by a factor of 100. Read the slope's units before calculating.

Reference: [1]

12. Update volatility over time

An exponentially weighted variance estimate blends the previous variance with the latest squared return or shock. Write the update as new variance = λ × prior variance + (1 − λ) × latest squared shock, using decimal returns. A λ near one preserves more of the earlier estimate; a lower λ responds faster to new information. Repeated updating gives recent shocks larger weights than distant ones because old contributions are multiplied by λ again each period. Take the square root only after updating the variance. The approach is useful when volatility changes over time, but its response depends on the chosen weight and on whether returns have been centered appropriately. A large shock raises the estimate quickly, while the estimate declines gradually if later shocks are small. It forecasts variability, not the direction of the next return.

Worked example: Assume prior daily variance is 0.0004, the latest squared decimal return is 0.0025, and λ = 0.94. New variance is 0.94(0.0004) + 0.06(0.0025) = 0.000376 + 0.000150 = 0.000526. Its square root is about 0.0229, or 2.29% daily volatility, above the prior √0.0004 = 2.00%.

Watch the distinction: A squared return of 0.0025 is an input to variance, not a 0.25% volatility forecast. Combine the variance terms first, then take the square root.

Reference: [1]

Financial Markets and Products

13. Value a forward payoff

A forward commits two parties to exchange an asset at a specified future date for an agreed delivery price. At expiry, the long party receives the asset and pays that price, so its payoff per unit is the asset's spot value minus the delivery price. The short party has the opposite payoff. This expiry payoff is distinct from the contract's value before expiry, which depends on current market terms and remaining time. A forward commonly has no exchange of its payoff until settlement, so gains may remain exposed to counterparty performance. State whether the question asks for payoff per unit, total payoff, or profit including any initial cost. Multiply a per-unit payoff by contract quantity only after finding the sign. FRM practice questions can reinforce this long-versus-short distinction.

Worked example: A hypothetical long forward covers one unit at a delivery price of $98. If the asset is worth $105 at expiry, payoff is $105 − $98 = $7: the buyer receives an asset worth more than the payment. If it is worth $94, payoff is $94 − $98 = −$4. For 100 units, those outcomes are a $700 gain or $400 loss.

Watch the distinction: The long forward does not gain whenever the asset price rises. Its expiry gain depends on whether the final spot price exceeds the agreed delivery price.

Reference: [1]

14. Follow futures margin

Futures positions are marked to market as the settlement price changes. For a long position, a price rise produces a daily cash credit and a price fall produces a daily cash debit through variation margin. Compute the change in settlement price, multiply by units per contract and the number of contracts, and preserve the sign. This daily cash movement differs from a forward's payoff that is typically settled at its contract date. Initial margin is collateral posted to support the position; it is not the purchase price of the underlying asset. A debit may lead to an additional margin call depending on the account balance and applicable arrangements, which cannot be inferred from price movement alone. Daily settlement also changes cash available for other obligations, making liquidity relevant even if the hedge is economically effective.

Worked example: A hypothetical trader is long 20 futures contracts, each covering 100 units. The settlement price falls from $50 to $48, a $2 decline per unit. Daily variation margin is 20 × 100 × (−$2) = −$4,000, so $4,000 is debited from the position. The underlying exposure is 2,000 units; the calculation does not imply delivery of those units today.

Watch the distinction: A $4,000 futures loss is a daily cash settlement, not merely an unrealized change waiting until expiry. Whether extra margin must be posted depends on the account balance.

Reference: [1]

15. Read a swap's cash flows

An interest rate swap exchanges interest amounts calculated on a reference notional. In a plain fixed-for-floating swap, one party receives a fixed rate and pays a floating rate for each payment period; the other party does the reverse. For a period, calculate each leg as notional × applicable annual rate × year fraction, then net receipts and payments from the stated party's perspective. The notional is generally a calculation base, not an amount exchanged in the periodic interest payment. A single period's net cash flow is not the swap's full present value, because future rates and discounting matter. Check the floating rate's reset period and the accrual fraction before calculating. A receiver of fixed benefits on that period's net payment when the fixed rate exceeds the applicable floating rate, though its overall value can change before payment.

Worked example: Assume a hypothetical $10 million notional, one-year accrual, and a party that receives 5% fixed while paying 4% floating. Fixed receipt is $10 million × 0.05 × 1 = $500,000. Floating payment is $10 million × 0.04 × 1 = $400,000. The net receipt is $100,000 for that period before discounting, with no exchange of the $10 million notional.

Watch the distinction: Multiplying the rate difference by notional gives a period cash flow only after applying the correct year fraction. It does not by itself value every remaining swap payment.

Reference: [1]

16. Separate option payoff from profit

A call gives its buyer the right to buy the underlying at the strike price. At expiry, its payoff per unit is the larger of the spot price minus the strike and zero. Profit for the buyer then subtracts the premium paid for the option. If the call expires out of the money, the buyer can decline exercise; the expiry payoff is zero, while the premium remains a loss. The writer's payoff and premium cash flow have opposite signs, although other costs can affect realized profit. Distinguish a payoff chart from a profit chart by locating the break-even price: for a call buyer at expiry, it is strike plus premium when ignoring financing and transaction costs. Before expiry, option value also reflects remaining time and uncertainty, so expiry payoff alone is not its current market price.

Worked example: A hypothetical call has strike $100 and premium $6 per unit. At expiry spot $115, payoff is max($115 − $100, 0) = $15, and buyer profit is $15 − $6 = $9. At spot $98, payoff is zero and profit is −$6. Ignoring other costs, break-even spot is $100 + $6 = $106.

Watch the distinction: Saying the buyer earns $15 at a $115 spot ignores the $6 premium. The $15 is exercise payoff; the stated buyer profit is $9 per unit.

Reference: [1]

17. Allocate securitization losses

A securitization divides cash flows and losses from an asset pool among tranches with different priority. Losses typically reach the most junior, first-loss tranche before they reach more senior tranches. To allocate a pool loss, subtract each tranche's loss-absorbing amount in order, passing any remainder to the next tranche. This ordering changes who bears a loss, but it does not remove the pool's underlying credit risk. Tranche value also depends on when losses occur, recoveries, cash-flow rules, and the dependence among borrowers: clustered defaults can rapidly exhaust junior protection. State the assumed waterfall and tranche sizes before calculating, since structures can differ. A senior tranche may experience no loss in a modest pool-loss scenario while still having exposure to a severe common shock.

Worked example: Suppose a hypothetical $100 million asset pool has a $5 million first-loss tranche and a next tranche large enough to absorb the remainder. A $7 million pool loss uses all $5 million of first-loss protection, leaving $7 million − $5 million = $2 million for the next tranche. The first tranche loses 100% of its $5 million principal in this simplified waterfall.

Watch the distinction: A 7% loss on the pool is not automatically a 7% loss on every tranche. Payment priority concentrates the first $5 million in the junior tranche.

Reference: [1]

18. Measure a repo haircut

In a repurchase transaction, one party raises cash against securities and agrees to repurchase them later. A haircut means the cash advanced is below the collateral's current market value. Calculate cash as collateral value × (1 − haircut), and the initial value cushion as collateral value minus cash. The cushion helps protect the cash lender if the borrower defaults and the collateral must be sold after its value falls. It is not a guarantee: a sufficiently large price decline, delayed sale, legal difficulty, or concentrated collateral can exceed it. Collateral may be revalued during the transaction, with additional collateral or cash required under the agreement. From the borrower's perspective, a larger haircut means less cash raised from the same securities, so haircuts also affect funding liquidity.

Worked example: Assume hypothetical collateral worth $10 million and a 5% haircut. Cash advanced is $10 million × (1 − 0.05) = $9.5 million, leaving a $0.5 million initial value cushion. If the collateral's value falls to $9.7 million, its marked value still exceeds the cash advanced by $0.2 million before sale costs and other claims.

Watch the distinction: The 5% haircut is measured against the $10 million collateral value, giving $0.5 million. It does not mean the lender can withstand every 5% market shock without loss.

Reference: [1]

Valuation and Risk Models

19. Discount promised bond cash flows

A bond's simple present value is the sum of each promised cash flow divided by a discount factor appropriate to its payment date and risk. For a fixed annual yield used across all maturities, discount a payment at year t by (1 + yield)^t. A coupon bond's final payment includes both the last coupon and principal, so omitting either produces a wrong price. If the coupon rate equals the stated yield under matching payment conventions, the price is at par. A higher yield lowers the present value of fixed promised payments because each denominator increases. This example prices promised flows; default risk, differing spot rates, accrued interest, and market quotation conventions may require further treatment. Keep the coupon and principal in the same currency unit, and state whether the yield is annual or periodic.

Worked example: A hypothetical two-year bond pays $5 at year one and $105 at year two, including $100 principal. At a 5% annual yield, price is $5/1.05 + $105/1.05² = $4.7619 + $95.2381 = $100. At the stated yield it sells at par; raising that yield would lower both present values and therefore the price.

Watch the distinction: Discounting only the $100 principal at maturity leaves out the final $5 coupon. The year-two cash flow is $105, while year one pays a separate $5.

Reference: [1]

20. Approximate rate-driven price changes

Modified duration estimates a bond's first-order percentage price response to a small change in yield: approximately minus duration times the yield change. Convexity adds a second-order term, one-half times convexity times the squared yield change. Use the yield change as a decimal, so a one-percentage-point rise is 0.01. Positive convexity usually makes the estimated loss from a yield rise smaller than duration alone suggests and the estimated gain from an equal yield fall larger. These are local approximations based on the measured bond and yield convention; they become less reliable for large moves or instruments whose cash flows change with rates. Check whether the task asks for a percentage price change or a currency change. For the latter, multiply the percentage estimate by the bond's starting price or position value.

Worked example: Suppose hypothetical modified duration is 4, convexity is 20 in compatible yield units, and yield rises by 0.01. Estimated ΔP/P is −4(0.01) + 0.5(20)(0.01²) = −0.04 + 0.001 = −0.039, or −3.9%. On a $1 million starting value, that is an approximate $39,000 loss; duration alone gives $40,000.

Watch the distinction: Entering a one-percentage-point yield move as 1 makes the estimate meaningless. Use 0.01, then interpret the result as a fraction of the starting price.

Reference: [1]

21. Price a one-step option

Price a one-step option by first writing its payoff in each possible stock-price state. A call pays the greater of the terminal stock price less the strike and zero. If you can buy shares and borrow cash to reproduce both payoffs, the option and that portfolio must have the same initial value under the model's assumptions. An equivalent shortcut uses a risk-neutral probability: choose a probability that makes the discounted expected stock price equal today's stock price, then discount the call's expected payoff. This is a pricing device, not a forecast of how often the stock will rise. Check that the implied probability lies between zero and one; otherwise the stated stock prices and financing assumptions permit an arbitrage. GARP's FRM exam topics place Valuation and Risk Models in Part I.

Worked example: Suppose a stock worth 100 will be 120 or 80 after one period, with no dividend or interest. A strike-100 call pays 20 or 0. The risk-neutral up probability is (100 − 80)/(120 − 80) = 0.5, so the call costs 0.5 × 20 = 10 per share. Equivalently, buy 0.5 share for 50 and borrow 40: the terminal values are 20 and 0.

Watch the distinction: A 0.5 risk-neutral probability does not mean the stock has a 50% real-world chance of rising. It is the probability implied by prices and the financing assumptions.

Reference: [1]

22. Hedge an option's first-order exposure

Hedge an option's first-order exposure by reading delta as the approximate change in option value for a one-unit change in the underlying price. For a call with delta 0.6, a one-unit stock rise increases the value of one long call by about 0.6 unit of currency, provided other inputs remain fixed and the move is small. A short call has the opposite exposure. Convert contract count into underlying units before choosing a stock hedge; contract multipliers can change the answer substantially. Then add stock with the opposite total delta. This neutralizes the local response to a small stock move, not the whole position's risk. Gamma changes delta as the stock moves, while time, volatility and financing conditions also affect option value. The FRM Part I study guide provides a related path through valuation topics.

Worked example: Assume each call covers one share. A trader is short 100 calls, each with delta +0.6 to its buyer. The option position has delta −100 × 0.6 = −60 share equivalents. Buying 60 shares adds +60, leaving net delta zero at the current price. A small stock rise should have roughly offsetting first-order effects; a large rise changes call delta and calls for reassessment.

Watch the distinction: Delta neutrality is not a permanent or complete hedge. It offsets the current first-order stock sensitivity, while gamma and other risk factors can still produce losses.

Reference: [1]

23. Estimate value at risk from scenarios

Estimate value at risk, or VaR, by defining a loss sign, a holding period, a confidence level and a quantile rule before sorting scenarios. For equally weighted observations ordered from smallest loss to largest, a selected upper quantile marks a loss level that most observations do not exceed. Quantile conventions differ, especially with small samples, so name the selected rank rather than assuming every method gives the same number. Historical scenarios preserve the joint market moves in the sampled dates, but their results depend on the chosen window and on how today's portfolio is repriced. VaR describes a boundary in the loss distribution. It does not report the average or maximum loss beyond that boundary, and an estimated historical quantile cannot guarantee the frequency of future breaches.

Worked example: Take 100 equally weighted one-day losses, sorted from smallest to largest, and define 95% VaR as observation 95. If that observation is 8 million currency units and observations 96 through 100 are larger, the estimated VaR is 8 million. Exactly five sampled losses exceed it. The estimate says nothing about whether those five losses are 9 million or 90 million.

Watch the distinction: A 95% VaR of 8 million is not a maximum possible loss. Losses above 8 million remain possible, and their magnitude requires separate analysis.

Reference: [1]

24. Measure losses beyond VaR

Expected shortfall describes the severity of losses in a specified tail. Start with the same loss horizon, portfolio and scenario weights used for VaR. Identify the worst fraction of outcomes and average their losses; with equal weights and a tail that contains a whole number of scenarios, this is an ordinary arithmetic mean. If a quantile cuts through an observation's probability weight, the convention for allocating that boundary weight matters. Expected shortfall responds to how large the worst losses become, whereas VaR can stay fixed when losses beyond its cutoff grow. It still inherits weaknesses in the scenarios: an omitted crisis mechanism cannot contribute to the estimate. Read the figure as a conditional tail average under the selected scenario distribution, not as the loss on every breach.

Worked example: Suppose the five worst of 100 equally weighted one-day scenarios, all above the chosen 95% VaR cutoff, lose 9, 10, 12, 15 and 24 million currency units. Their total loss is 70 million, so expected shortfall for this five-scenario tail is 70/5 = 14 million. The 24 million observation raises that average even if the VaR cutoff remains unchanged.

Watch the distinction: Expected shortfall is not found by adding VaR to the average excess above VaR unless the quantities are defined consistently. Average the selected tail losses themselves.

Reference: [1]

Market Risk Measurement and Management

25. Map positions to market factors

Map a position to the prices and rates that actually change its value. A foreign-currency bond, for example, responds to its local bond price and to the exchange rate used to convert proceeds into the investor's currency. The bond price can itself move with benchmark yields, credit spreads and optionality. Choose a clear quotation for the exchange rate, then write the investor-currency value as local value multiplied by investor-currency units per foreign-currency unit. For percentage changes, multiply the gross returns rather than simply adding them; the product includes their interaction. This factor map guides hedging and scenario design because an apparently successful local bond trade can still lose money after conversion. FRM Part II study material connects this valuation work to market risk measurement.

Worked example: Assume a foreign bond starts at 100 in local currency and the exchange rate is 1 investor-currency unit per local unit. Its local price rises 2% to 102, while each local unit buys 3% less investor currency, or 0.97. Final converted value is 102 × 0.97 = 98.94, versus 100 initially. The investor's return is 98.94/100 − 1 = −1.06%.

Watch the distinction: Adding a 2% bond gain and a 3% currency loss gives only an approximation. The exact converted return includes their product and equals −1.06% here.

Reference: [1]

26. Reprice under historical moves

Historical scenario analysis applies a past day's joint factor changes to today's portfolio. Begin with current positions and a consistent set of factor exposures. For each historical date, move all relevant prices, rates and exchange rates together, then revalue the current holdings. Keeping the moves paired preserves dependence observed on that date, such as an equity decline occurring alongside a currency move. A simple dollar-sensitivity calculation can illustrate the method, but options and other nonlinear positions require full repricing to capture curvature. The resulting profit and loss belongs to today's holdings under a past market move; it is not the portfolio's actual profit and loss on that historical date. Check whether the chosen history contains stresses relevant to the current positions and whether changed market structure limits its usefulness.

Worked example: Today's portfolio holds 10 million currency units of equity and has a separate long currency exposure worth 5 million in reporting currency. On one past date, equity fell 4% and that currency rose 2%. Applying both moves gives −0.04 × 10 million + 0.02 × 5 million = −400,000 + 100,000 = −300,000, assuming linear exposures and no other factors.

Watch the distinction: Do not apply the past percentage moves to past position sizes. The purpose is to ask how today's holdings would respond to those jointly observed moves.

Reference: [1]

27. Use sensitivity and stress together

A sensitivity measures the local change in value when one factor moves slightly and other factors are held fixed. It is useful for quick comparisons, hedge sizing and identifying dominant exposures. To estimate a larger shock, multiplying the local sensitivity by the shock size assumes that sensitivity stays constant. A stress calculation instead revalues the position at the shocked inputs, allowing curvature and changing cash-flow behavior to appear. Specify whether the stress moves one factor or several together: a difference between a local estimate and full repricing can reflect curvature only when the other assumptions match. Use both views. The local measure tells you the immediate slope; stress shows the loss under a particular adverse state. Neither alone describes all possible market paths or funding effects.

Worked example: Suppose a bond position loses 40,000 currency units for a small one-basis-point yield rise. Linear extrapolation to a 100-basis-point rise gives 100 × 40,000 = 4 million of loss. Full repricing with only that yield changed shows a 3.6 million loss. The 0.4 million difference indicates that the local slope did not remain constant over the larger move.

Watch the distinction: A one-basis-point sensitivity cannot automatically be multiplied into an accurate large-shock loss. Repricing can reveal curvature, and a multi-factor stress may add effects beyond curvature.

Reference: [1]

28. Investigate VaR exceptions

A VaR exception occurs when a comparable realized loss exceeds the VaR forecast for the same horizon. First align the dates, currencies, position set and profit-and-loss definition. A forecast based on yesterday's holdings is most cleanly assessed against profit and loss from holding those positions through today's market moves; intraday trading, fees and valuation adjustments can obscure that comparison. Record the exception, then investigate its cause. An unusual move may be a legitimate tail event, while repeated breaches may expose stale data, missing factors, changing dependence or poor model calibration. The amount above VaR describes this breach but does not alone prove a model failure. FRM practice questions can help you rehearse the comparison between a forecast and an observed loss.

Worked example: A one-day VaR forecast is 2 million currency units. Comparable one-day profit and loss on the forecast portfolio is −2.5 million, a loss of 2.5 million. Because 2.5 exceeds 2, record one exception; its exceedance is 2.5 − 2 = 0.5 million. Review that day's market moves and any position or valuation changes before interpreting the exception.

Watch the distinction: One exception does not establish that a VaR model is defective. Tail losses are possible by design; the definition of profit and loss and the pattern of breaches matter.

Reference: [1]

29. Match interest-rate sensitivity

Dollar duration translates a small yield move into an approximate currency-value change. State its unit, such as currency units per basis point, and its sign: a conventional long bond usually loses value when yield rises. To size a hedge, divide the portfolio's sensitivity magnitude by the opposite sensitivity supplied by one hedge contract. Choose the contract direction so its gain offsets the portfolio's loss for the same yield move. Then check what rate each position tracks. Matching sensitivity to a parallel shift does not eliminate risk from a change in curve shape, different reference rates, credit spreads or an imperfect hedge instrument. Duration is also local, so a large yield shock calls for repricing and possibly a convexity check. The calculation sizes an initial hedge, not a guarantee of zero profit and loss.

Worked example: A bond portfolio loses approximately 200,000 currency units per basis point when yields rise. Assume one appropriately directed hedge contract gains 5,000 per basis point for the same parallel move. The required size is 200,000/5,000 = 40 contracts. For a small one-basis-point rise, the estimated bond loss of 200,000 is offset by 40 × 5,000 = 200,000 of hedge gain.

Watch the distinction: Dividing dollar durations gives a contract count only after their units and signs are aligned. A matched parallel-rate sensitivity can still leave curve and spread risk.

Reference: [1]

30. Recognize basis risk

Basis risk is the residual risk that appears when the price of a hedged exposure and its hedge do not move together as assumed. A hedge may use a related bond, index, futures contract or currency pair because the exact exposure cannot be traded conveniently. Its effectiveness then depends on the relationship between the two prices, including contract terms, quality, location and maturity. Measure the combined profit and loss and inspect the change in the relevant price spread, or basis. Even a correctly sized hedge can leave a loss when that spread changes. Distinguish this from a simple sizing mistake: if the hedge had the intended sensitivity to its own underlying but that underlying diverged from the asset, the remaining loss is basis risk. Stress scenarios should include plausible basis widening.

Worked example: A firm hedges an asset with a related instrument. Over a week, the asset loses 1.2 million currency units and the hedge gains 1.0 million. Combined profit and loss is −1.2 + 1.0 = −0.2 million, a 200,000 loss. If the positions were sized as intended, investigate the unmatched 200,000 as a change in their price relationship.

Watch the distinction: A hedge that gains money has not necessarily worked completely. Its gain must be compared with the exposure's loss, using the same period, currency and position size.

Reference: [1]

Credit Risk Measurement and Management

31. Calculate expected credit loss

Expected credit loss combines three questions over a stated horizon: how likely is default, how much is owed at default, and what fraction of that exposure would be lost after recovery? Write these as probability of default (PD), exposure at default (EAD) and loss given default (LGD). In a simple one-period setting, expected loss equals PD × EAD × LGD. PD and EAD must refer to the same horizon, while LGD should reflect the assumed recovery and any relevant costs. A credit line can have greater EAD than today's balance if the borrower draws before default. Expected loss is a probability-weighted average across outcomes, not a prediction that a particular borrower will make a partial payment. Portfolio tail loss also depends on default dependence and cannot be inferred from this mean alone.

Worked example: Assume a one-year PD of 2%, EAD of 10 million currency units at default, and LGD of 40% after recovery. Conditional loss on default is 0.40 × 10 million = 4 million. The one-year expected loss is 0.02 × 4 million = 80,000. This is an average across many comparable risk outcomes, not an invoice for 80,000 from this borrower.

Watch the distinction: Multiplying PD by the current loan balance can understate loss if the balance may grow before default. Use exposure at default and keep the horizon consistent.

Reference: [1]

32. Spot credit concentration

Credit concentration can hide behind a count of different borrower names. Ask what economic shock affects each borrower and whether one shock could cause several defaults together. Shared commodity prices, regions, customers or funding sources can make defaults dependent. Assess both the exposure to each name and the size of a plausible joint loss after recovery. A portfolio with two equal loans is diversified by legal borrower, but may be concentrated in one underlying factor. Expected losses can be added across names, yet the chance of a severe combined loss depends on their default dependence. A useful decision sequence is to identify the common driver, estimate exposures and recoveries under that driver, and compare the joint outcome with available risk capacity. Independence should be an examined assumption, not a default conclusion.

Worked example: Two borrowers each owe 10 million currency units and rely on the same commodity price. If a price collapse causes both to default and each debt recovers 30%, recovery is 3 million per borrower. Each loss is 10 − 3 = 7 million; combined loss is 14 million. The calculation illustrates a joint stress, without assigning it a probability.

Watch the distinction: Two borrower names do not imply two independent risks. A shared revenue driver can make their defaults coincide and produce a much larger portfolio tail loss.

Reference: [1]

33. Interpret default-model mechanisms

A structural default model links default to the firm's asset value relative to promised debt at a specified payment date. If assets are insufficient to pay due debt, equity holders have no residual claim and creditors face a shortfall, subject to recovery details. This mechanism connects leverage and asset volatility to credit risk, but the firm's asset value is usually not directly observed. An intensity model begins differently: it specifies a conditional rate at which default can occur through time. With a constant annual intensity λ in a simple model, survival to time T is exp(−λT), so default probability by T is 1 − exp(−λT). Intensity need not explain which balance-sheet event causes default. Use the structural view to reason about the asset cushion; use the intensity view to model default timing and pricing.

Worked example: At maturity, suppose a firm has 90 million currency units of assets and owes 100 million. In a simplified structural illustration with no other claims or costs, creditors receive 90 million and face a 10 million shortfall; equity receives zero. An intensity model would instead start with a specified default intensity to estimate the probability of default over that period, rather than deriving this shortfall from asset value.

Watch the distinction: A 10 million asset shortfall is a loss amount in one maturity state, not a default probability. An intensity describes default arrival, not the creditor's recovery by itself.

Reference: [1]

34. Value a credit rating migration

A credit rating migration can change a bond's market value before any payment is missed. A downgrade may cause investors to demand a wider credit spread over the benchmark yield. Discounting the bond's remaining cash flows at a higher required yield lowers its price, even if the issuer continues paying coupons. To interpret the move, hold other drivers conceptually separate: benchmark-rate changes and spread changes can both affect the same bond. A rating is a coarse description of credit quality; market spreads may move before or without a rating action. For a held bond, a mark-to-market loss is the change in its measured value under the valuation basis used. It should not be confused with realized cash loss from default, nor should a downgrade automatically be treated as default.

Worked example: Assume a bond is quoted at 100 currency units per 100 of face value before a hypothetical downgrade. Its credit spread widens while the benchmark yield and promised payments stay fixed, and the bond reprices to 94. The holder's mark-to-market change is 94 − 100 = −6 per 100 of face value. The issuer may still pay every scheduled coupon.

Watch the distinction: A downgrade need not wait for missed payment to create a loss. The price can fall because investors demand greater compensation for perceived future credit risk.

Reference: [1]

35. Link exposure to CVA

Credit valuation adjustment, or CVA, reflects the possibility that a counterparty defaults while a derivative has positive value to you. At that point, the positive value is an exposure: replacement may cost money and recovery may be incomplete. A simplified calculation combines expected positive exposure at a future date, the probability of default in that interval and LGD, then discounts the expected loss. A fuller calculation considers exposures across dates, survival, netting, collateral and the relationship between exposure and default. If exposure tends to rise as the counterparty weakens, multiplying an unconditional average exposure by default probability can understate the adjustment. Keep signs clear: counterparty credit risk generally reduces the value of a positive derivative claim to the holder. CVA is a valuation adjustment, not a cash payment already lost.

Worked example: For a single future period, assume expected positive exposure is 5 million currency units, default probability over that period is 2%, and LGD is 60%. Assume exposure and default are independent and ignore discounting. Approximate CVA is 5 million × 0.02 × 0.60 = 60,000. Under these assumptions, counterparty risk reduces the claim's value by approximately 60,000.

Watch the distinction: Current derivative value is not automatically the exposure used for CVA. Future positive exposure changes with markets, and its relationship with default can materially affect the result.

Reference: [1]

36. Account for collateral and wrong-way risk

Collateral reduces counterparty exposure only to the extent that it is available and eligible when needed. Start with the derivative's positive replacement value, apply enforceable netting where relevant, then subtract the recognized value of collateral. Haircuts, currency mismatches, settlement timing and disputes can make recognized collateral worth less than its quoted market value. The result is current residual exposure; future exposure can rise before another margin exchange. Wrong-way risk adds a dependence problem: the derivative's value to you may grow precisely when the counterparty becomes less able to pay. A hedge tied to the counterparty's own credit or business conditions can create this pattern. Diagnose it by asking what event increases the claim, what the same event does to counterparty strength, and whether collateral remains reliable in that event.

Worked example: A derivative has a positive replacement value of 8 million currency units. Assume 5 million of posted collateral is fully eligible, available and recognized, with no other netting. Current residual exposure is max(8 − 5, 0) = 3 million. If the derivative gains value when the counterparty's credit quality deteriorates, future exposure and default risk may rise together, creating wrong-way risk.

Watch the distinction: Five million of collateral does not erase an eight million claim. Nor does today's 3 million residual describe future exposure if markets move or collateral loses value.

Reference: [1]

Operational Risk and Resilience

37. Classify an operational loss

Classify an operational loss by tracing the initiating failure and its consequence. The cause may involve a person, process, system or external event; several can interact. A wrong payment instruction, for example, is a process or execution failure even if the resulting cash loss resembles a credit exposure. Record the gross amount sent incorrectly, any recovered amount and the remaining direct loss separately. This avoids treating a temporary misdirection as a permanent loss or hiding an unrecovered amount behind eventual recovery efforts. Other consequences, such as customer disruption or reputational damage, should be described on their own terms rather than forced into the direct cash-loss figure. For control design, follow the payment from instruction creation through authorization, transmission and reconciliation to find where detection or prevention could have occurred.

Worked example: An incorrect payment instruction sends 200,000 currency units to the wrong account. The bank later recovers 150,000 and assumes the remaining 50,000 is unrecoverable. Gross misdirected cash was 200,000; the direct unrecovered cash loss is 200,000 − 150,000 = 50,000. Classify the initiating event as a payment-process failure and examine instruction checks and reconciliation for control gaps.

Watch the distinction: The amount initially sent to the wrong account is not necessarily the final loss. Track recoveries and record any customer or reputational effects separately from unrecovered cash.

Reference: [1]

38. Assess controls against a scenario

A risk and control assessment starts with a specific adverse event, then asks how it could occur and what each control does to interrupt it. Inherent exposure describes the scenario before the listed controls; residual exposure considers their effectiveness. For a duplicate payment, a maker-checker review can challenge the instruction before release, while an automated duplicate flag can compare payment fields against earlier instructions. These controls address overlapping but different failure routes. Test their design and operation: which fields are matched, who can override a warning, and whether reviewers have enough information and time. A potential loss amount alone does not measure frequency or residual risk. If controls reduce the chance of an event, do not claim they reduce its severity unless they also limit the amount paid or improve recovery.

Worked example: Suppose an accidental duplicate payment could create a 500,000 currency-unit loss if unrecovered. Before controls, 500,000 is the stated loss severity for that event. A maker-checker review may catch a repeated instruction before approval; an automated flag may catch a matching payment before release. If both fail, the potential unrecovered loss remains 500,000. No residual expected loss can be calculated without event and control-failure probabilities.

Watch the distinction: Listing two controls does not prove the scenario is safe. Both may rely on the same incorrect payment data, or reviewers may routinely override the automated warning.

Reference: [1]

39. Choose an actionable risk indicator

A key risk indicator is useful when its definition is stable, its movement can be interpreted and someone can act on it. A failed-settlement rate divides failed settlements by the relevant number of trades over the same period. Keep counting rules consistent: a change in which trades qualify as failures can move the indicator without any change in the underlying process. Compare both the level and the change, then investigate possible causes such as staffing, system performance, trade complexity or counterparties. The indicator signals where to look; it does not by itself identify the cause or measure the cash loss. A response can include reviewing failed trades by source, checking whether delays are concentrated in one system or counterparty, and verifying that reconciliation catches unresolved items promptly.

Worked example: In one period, 8 of 1,000 trades fail to settle, a rate of 8/1,000 = 0.8%. In the next, 20 of 1,000 fail, or 2.0%. The increase is 1.2 percentage points, and the rate is 2.0/0.8 = 2.5 times its earlier level. Break down the 20 failures by system, team and counterparty to guide a control response.

Watch the distinction: A higher failed-settlement rate does not prove a particular employee or system caused it. The denominator, trade mix and failure definition must be checked before assigning a cause.

Reference: [1]

40. Combine loss data with scenarios

Internal loss observations show what the organization has experienced and how its controls behaved, but rare severe events may be absent from a short record. Scenarios can add plausible losses that have not occurred internally, provided their assumptions about frequency, severity and dependencies are explicit. Use both sources to challenge each other. A quiet loss history should prompt questions about exposure and reporting, while an extreme scenario should be tested against the business process and available evidence. A simple frequency-times-severity product gives an annual average under a stated event model. It does not describe the distribution's tail, the possibility of multiple events, or uncertainty in the frequency estimate. Keep the severity of an event separate from its probability so that a low annual average does not conceal an intolerable single loss.

Worked example: Assume a hypothetical operational event occurs in a given year with probability 0.1 and causes a 4 million currency-unit loss when it occurs. If the annual outcome is either no event or one event, expected annual loss is 0.1 × 4 million + 0.9 × 0 = 400,000. The 4 million event remains possible despite the much smaller average.

Watch the distinction: An expected annual loss of 400,000 is not a cap on annual loss. It is an average under the assumed event probability and severity, both of which may be uncertain.

Reference: [1]

41. Plan for service disruption

A critical service is an outcome the firm must deliver, such as completing customer payments. Map the people, data, systems, connections, and outside providers needed to deliver it. Then ask what fails if each dependency stops and whether an alternative can operate in time. Recovery order follows the dependency chain: restoring an application cannot restart payments while its required data feed is unavailable. The sequence should also protect data integrity. After service resumes, reconcile queued instructions against completed payments so that retries do not create duplicates or omissions. Test the plan with realistic interruptions, including a failure during a busy period, and record where handoffs or capacity break down. GARP's FRM program lists Operational Risk and Resilience among its Part II domains.

Worked example: A payment service needs an incoming data feed, a processing application, and a bank connection. The application and connection work, but the feed fails. Restore the feed first, then verify that the application receives complete records. If 500 payments were queued, reconcile all 500 against bank acknowledgments before releasing retries. That sequence restores the missing dependency and checks that recovery has not created duplicate payments.

Watch the distinction: Restarting every component at once may look faster, but it can obscure the failure and duplicate queued transactions. Restore the broken dependency, then verify the complete service.

Reference: [1]

42. Challenge model use

A model developer chooses data, features, methods, and intended uses. Independent validation challenges those choices before reliance on the model, checking conceptual soundness, data quality, implementation, and performance across relevant segments. Monitoring continues after deployment because customers, transaction sizes, incentives, and fraud patterns change. Compare predictions with observed outcomes by segment, not only in aggregate: strong overall accuracy can conceal poor detection where losses are largest. When performance deteriorates, first check for data errors and changes in the population; then measure the business effect. Responses may include tighter human review, limits on model use, recalibration, or redevelopment. Keep a record of the change and test the revised model before expanding its use. The Associate in Risk Management guide offers related control concepts.

Worked example: A fraud model was trained mainly on small transfers. In a later sample, it flags 90 of 100 fraudulent small transfers but only 10 of 40 fraudulent large transfers. Detection is 90% versus 25%, respectively. The size split reveals a material weakness that an overall rate of 100 out of 140, or 71.4%, hides. Route large transfers to additional review while the model is investigated.

Watch the distinction: A model's accuracy at launch does not establish continuing reliability. A changing transaction mix can expose weak segments even when an aggregate performance figure appears acceptable.

Reference: [1]

Liquidity and Treasury Risk Measurement and Management

43. Separate funding and market liquidity

Funding liquidity concerns the ability to meet cash obligations when due. Market liquidity concerns the ability to trade an asset promptly at a price close to a reasonable valuation. A firm may have valuable assets yet lack cash today; selling those assets into a thin market can turn the funding problem into a trading loss. Diagnose the cash amount and deadline first. Next, identify assets or borrowing sources that can produce usable cash by that deadline, allowing for settlement time, haircuts, and transaction costs. Compare executable prices with current marks, while recognizing that a mark may itself be stale. If selling an asset depresses its price, the firm may need to sell more than expected. The two liquidity risks therefore reinforce each other during stress.

Worked example: A firm owes 20 million today and has no other cash. An asset marked at 20 million can be sold today for only 18 million, with immediate settlement. The sale produces 18 million, leaving a 2 million funding gap. Relative to the mark, the executable sale price is also 2 million lower. The firm needs another timely cash source; holding the asset does not pay today's obligation.

Watch the distinction: An asset's marked value is not necessarily cash available today. The executable price, sale capacity, and settlement time determine how much it can contribute to funding.

Reference: [1]

44. Construct a cash-flow ladder

A cash-flow ladder places expected inflows and outflows into time buckets, then carries each bucket's closing cash into the next. For bucket t, closing cash equals opening cash plus usable inflows minus required outflows. The first negative closing balance identifies when funding is needed, while the most negative cumulative balance shows the largest projected shortfall under those assumptions. Timing matters as much as totals: an inflow next month cannot meet a payment this week. Distinguish contractual cash flows from uncertain ones, such as anticipated loan repayments or asset sales, and consider stress cases where the latter arrive late. Check currency and legal entity as well; cash held in one place may not be transferable to another in time. A ladder turns a broad liquidity concern into a dated funding decision.

Worked example: Opening cash is 30 million. During the week, usable inflows total 15 million and required outflows total 50 million. Closing cash is 30 + 15 − 50 = −5 million. If all amounts use the same currency and the inflows arrive before the payments, at least 5 million of additional funding is needed that week. Later receipts cannot remove that week's gap.

Watch the distinction: Netting all future inflows against all future outflows hides timing. A positive monthly total does not prevent a shortfall in an earlier daily or weekly bucket.

Reference: [1]

45. Estimate buffer survival

Buffer survival estimates how long a specified stock of usable liquid resources can cover a specified net cash outflow rate. Divide the buffer by the burn rate, keeping units consistent. The result is a planning horizon under stated assumptions, not a regulatory ratio or a promise that the firm can operate for that long. Check whether assets can be monetized when needed and whether cash is in the right currency and entity. A constant daily burn is convenient for a simple calculation, but real withdrawals and margin calls may accelerate. Recalculate survival for faster outflows, lower sale proceeds, and delayed inflows. A useful decision sequence is to identify immediately usable cash, estimate dated net outflows, locate the first uncovered day, and prepare funding actions before that day arrives.

Worked example: A treasury team has a hypothetical 120 million usable buffer and projects net outflows of 30 million per day. Assuming the rate remains constant and all buffer assets produce cash when needed, survival is 120 million ÷ 30 million per day = four days. At 40 million per day, it falls to three days. These are scenario results, not minimum required horizons.

Watch the distinction: Dividing total assets by daily outflows overstates survival if some assets are pledged, slow to sell, or unavailable to the entity facing payments.

Reference: [1]

46. Stress funding assumptions

A liquidity forecast depends on assumptions about customer payments, maturing assets, funding renewals, and collateral calls. Stress testing changes those assumptions together and recalculates the dated cash position. An inflow that arrives late and a larger margin call both reduce cash available on the payment date, even though they arise from different channels. Start with the base cash ladder, change each affected line, and preserve the timing of every change. Then check whether proposed responses, such as drawing a line or selling collateral, remain available in the same scenario. Avoid counting an asset sale twice as both a planned inflow and a contingency source. Compare the stressed gap with usable buffers and committed funding to decide which actions must begin before the gap occurs.

Worked example: A base forecast includes 40 million of inflows and a 10 million margin call. Under stress, inflows are 25 million and the call is 18 million. Relative to base, delayed or missing inflows reduce cash by 40 − 25 = 15 million; extra margin uses 18 − 10 = 8 million. The cash position worsens by 23 million, assuming all other flows are unchanged.

Watch the distinction: Changing only one forecast line can miss a joint squeeze. A market shock may delay inflows while simultaneously increasing collateral demands.

Reference: [1]

47. Price funding into business decisions

Funds transfer pricing assigns a funding cost to a product or desk so its reported margin reflects the resources it uses. A loan that ties up funds for longer should be assessed using a charge consistent with that funding need, rather than a convenient short-term rate. Begin with the product's yield, subtract expected credit and operating costs, and subtract the internal funding charge on a comparable annual basis. This is a decision measure, not necessarily the loan's accounting profit: capital costs, taxes, fees, and unexpected losses may also matter. If the funding charge rises while the customer yield stays fixed, the apparent margin narrows. That signal can change pricing, loan term, or volume decisions and helps avoid rewarding activity that consumes scarce stable funding.

Worked example: A hypothetical loan earns 6% annually and has expected credit and operating costs totaling 2% annually. With a 3% internal funding charge, its simplified margin is 6% − 2% − 3% = 1% per year. At a 2% charge, it would be 2%. The one-percentage-point difference reflects the assumed funding cost, before other costs and income.

Watch the distinction: Using the cheapest overnight funding rate for a long loan can exaggerate its return. The assigned charge should reflect the product's funding profile and assumptions.

Reference: [1]

48. Sequence contingency funding

A contingency funding plan ranks cash sources by when they become usable, how reliable they are under stress, and what using them does to future liquidity. Immediately available cash usually addresses the earliest payment. A committed line may bridge a remaining gap, but its terms, currency, draw process, and counterparty availability must be checked. Asset sales can supply cash only after execution and settlement; selling into stress may also impose losses or weaken collateral capacity. Sequence actions against the cash-flow ladder rather than merely listing possible sources. Reserve time for approvals, transfers, and operational steps. After covering today's obligations, update the ladder for tomorrow: a line draw creates future repayment needs, and spending the entire cash reserve removes protection against another shock.

Worked example: A firm faces a 25 million payment gap today. It has 10 million available cash and can draw 15 million today from a committed line whose conditions it meets. The two sources provide 25 million, exactly covering the gap. A separate asset sale would settle tomorrow, so it contributes zero to today's payment. Treasury must also plan for the line's later repayment.

Watch the distinction: A source that appears large enough on paper may arrive too late. Match its usable cash date to the obligation date before counting it.

Reference: [1]

Risk Management and Investment Management

49. Relate beta to required return

The capital asset pricing model links an asset's required return to its sensitivity to broad market movements. Under the model's assumptions, required return equals the risk-free rate plus beta times the expected market risk premium. Beta measures how much the asset's return tends to move with the market in a linear relationship; it does not measure all sources of risk. A beta above one implies greater market sensitivity than the market portfolio. Use rates for the same horizon and currency, and distinguish an expected required return from a realized return. The model's output can be compared with a forecast return to assess whether compensation appears adequate, but the conclusion depends on the beta estimate, premium estimate, and model assumptions. Firm-specific shocks can still produce large losses.

Worked example: Suppose the one-year risk-free rate is 3%, the expected one-year market premium is 5%, and a stock's beta is 1.2. Its model-implied required return is 3% + 1.2 × 5% = 9% for that year. The 6% premium above the risk-free rate compensates for modeled market exposure; a 9% return is not guaranteed.

Watch the distinction: Beta 1.2 does not mean the asset will earn 20% more than the market. It describes estimated market sensitivity within the model.

Reference: [1]

50. Measure active-return variability

Tracking error measures the variability of returns relative to a benchmark. For each period, subtract the benchmark return from the portfolio return to obtain active return. Then calculate the standard deviation of those active returns using a stated population or sample convention. A portfolio can have a small average active return but large tracking error if it alternates between substantial wins and losses. Conversely, a stable positive active return has low tracking error over the observed periods. Match observation frequency before comparing figures, and annualize only with an explicit assumption about how period returns relate over time. Tracking error describes dispersion around average active return; it does not show whether the manager adds value or whether the benchmark fits the mandate. FRM practice questions can reinforce the calculation.

Worked example: For two periods, active returns are +2% and −2%. Their mean is (2% − 2%) ÷ 2 = 0%. Treating these two observations as the full population, variance is [(2 − 0)² + (−2 − 0)²] ÷ 2 = 4 percentage-points squared. Tracking error is √4 = 2% per period; no annual rate follows without further assumptions.

Watch the distinction: Average active return is not tracking error. Gains and losses can cancel in the average while still showing substantial variability around the benchmark.

Reference: [1]

51. Allocate portfolio risk

Portfolio volatility depends on each position's weight, its own volatility, and its covariance with the rest of the portfolio. Marginal risk contribution asks how total volatility would change after a small increase in a position's weight, with the measurement convention held fixed. Component risk contribution multiplies that marginal effect by the existing weight; the components sum to portfolio volatility for the usual homogeneous volatility measure. A position with high standalone volatility may contribute little if it offsets other holdings. Conversely, several modest positions can collectively dominate risk when they share a factor. To decide which trade reduces total risk most, inspect contributions and correlations, then recalculate after the proposed trade. Contributions change with weights and market conditions, so a current ranking is not permanent.

Worked example: Two assets each have 10% volatility, equal 0.5 weights, and zero correlation. Portfolio variance is 0.5² × 10² + 0.5² × 10² = 50 percentage-points squared, so volatility is √50 = 7.07%. Symmetry assigns half that risk to each asset: 3.54 percentage points each, allowing for rounding. Their contributions sum to approximately 7.07%.

Watch the distinction: Adding standalone volatilities gives 10% for this equally weighted pair, but it ignores diversification. Component contributions must account for covariance with the whole portfolio.

Reference: [1]

52. Find hidden factor exposures

Different securities or funds can carry the same underlying exposure. Look through each holding to the market factors that drive its gains and losses, such as interest rates, equity markets, currencies, or credit spreads. Express sensitivities in comparable units and aggregate positions with the same factor, allowing negative exposures to offset positive ones only when their responses are genuinely comparable. Names and asset classes alone do not establish diversification: an equity fund and a convertible bond fund may both lose when equities fall. Factor estimates are approximations, especially for options and assets whose sensitivity changes in stress. After aggregating, apply plausible factor moves and inspect residual risks that the simple mapping leaves out. This reveals concentrations that a count of holdings would miss.

Worked example: Fund A and Fund B each have a hypothetical 5 million positive exposure to the same equity factor, measured as value that moves one-for-one with that factor for a small shock. Combined exposure is 10 million. A 4% factor decline implies approximately 10 million × −0.04 = −400,000. Fund labels do not reduce this shared loss; other factors could alter the final result.

Watch the distinction: Owning two funds does not automatically halve risk. If both funds respond to the same shock, their factor exposures add despite their different names.

Reference: [1]

53. Calculate drawdown

Drawdown measures how far portfolio value falls below its previous high. At each point, track the highest value reached so far; current drawdown equals (peak − current value) divided by that peak. Maximum drawdown is the largest such decline over the observed path. It depends on the order of returns: the same set of periodic returns can produce different peaks and troughs when rearranged. Drawdown captures a sustained loss from a prior high, which one-period volatility does not describe. It is backward-looking and sensitive to the chosen observation window and valuation frequency. When comparing managers, also consider how long recovery takes and whether infrequently valued holdings smooth the path. A past maximum is a description of history, not a bound on future losses.

Worked example: A portfolio's observed values move from 100 to 120, then 110, then 90, and finally 115. The running peak after reaching 120 stays at 120. Drawdown at 90 is (120 − 90) ÷ 120 = 25%, the largest decline on this stated path. The later value of 115 remains about 4.17% below the 120 peak, so the portfolio has not fully recovered.

Watch the distinction: The fall from 100 to 90 is 10%, but 100 is not the relevant peak. Drawdown uses the highest earlier value, which is 120 here.

Reference: [1]

54. Attribute active performance

Active performance is portfolio return minus benchmark return. Attribution asks which decisions produced that difference. In a simple sector allocation calculation, compare the portfolio's sector weight with the benchmark's weight, then multiply that weight difference by the sector benchmark return minus the overall benchmark return. An overweight helps when that sector outperforms the overall benchmark and hurts when it underperforms. The result is a contribution in percentage points, not the portfolio's total active return. Security selection within sectors, interactions between weights and security returns, fees, and other positions can add or subtract from it. State the attribution convention before comparing reports because formulas differ. Use the calculation to diagnose a decision, then check whether the observed result reflects a persistent exposure or a temporary market move.

Worked example: A portfolio allocates 30% to a sector while its benchmark allocates 20%. Suppose that sector's benchmark return exceeds the overall benchmark return by 4 percentage points. The simplified allocation contribution is (0.30 − 0.20) × 4% = 0.4 percentage points of active return. This identifies the benefit of the overweight under the stated convention; total active return also depends on other decisions.

Watch the distinction: A positive allocation contribution is not the portfolio's total excess return. Selection, other sector weights, and interaction effects may offset or amplify it.

Reference: [1]

Current Issues in Financial Markets

55. Translate climate scenarios into exposure

A climate scenario becomes financially useful when its shock is translated into borrower and collateral cash flows. A physical event may damage assets or interrupt operations; a transition change may alter energy costs, demand, or investment needs. Identify the affected location or business line, estimate how revenue and costs change, and carry that change into cash available for debt service. Then assess whether the borrower can refinance, provide collateral, or absorb the shock. Collateral values may move at the same time as borrower strength, increasing loss severity if default occurs. Scenario results depend on timing, adaptation, insurance, and assumptions about prices. They are conditional estimates, not predictions. GARP's FRM program lists Current Issues in Financial Markets as a Part II domain.

Worked example: A hypothetical policy cost reduces a borrower's annual cash available for debt service from 12 million to 8 million, while scheduled annual debt service remains 10 million. Coverage falls from 12 ÷ 10 = 1.2 to 8 ÷ 10 = 0.8. The new cash flow is 2 million short of scheduled service before any reserves or refinancing, prompting closer examination of liquidity and default risk.

Watch the distinction: A climate label alone does not quantify credit risk. Trace the shock through the borrower's costs, cash flow, debt service, and collateral before drawing conclusions.

Reference: [1]

56. Map shared technology dependencies

A common technology provider can create correlated disruption across institutions that appear otherwise independent. Map which services each firm receives, what share of activity depends on them, and whether alternative routes have enough capacity. The provider may support authentication, cloud hosting, market data, or payment messaging; failure at one point can stop several downstream processes. Estimate affected volume separately for each firm, then inspect system-wide consequences such as delayed settlements and simultaneous demand for backup capacity. The percentages cannot simply be added into a single firm's exposure, and having three customers does not diversify a shared outage. Assess recovery by tracing dependencies and testing failover under a scenario where many clients switch at once. The Part II study guide connects this issue with broader risk topics.

Worked example: Three firms each route 40% of their own payment volume through one provider. If the provider fails completely and no backup operates, each firm's exposed share is 40%, not 120%. If their daily volumes are 10, 20, and 30 million, affected payments are 4, 8, and 12 million, totaling 24 million across the firms. The outage creates simultaneous exposure, not diversification.

Watch the distinction: Multiple customers do not make a common provider's failure independent across firms. The shared dependency can interrupt all affected payment streams at the same time.

Reference: [1]

57. Monitor AI-driven model change

An automated lending model can change financial risk after deployment even if its code stays fixed. New borrower segments, altered data sources, and feedback from automated approval decisions can shift the population the model sees. Monitor predicted default rates against observed outcomes by segment and vintage, while checking whether the observation period is long enough for defaults to emerge. A gap may signal data defects, changed borrower behavior, or a model that no longer ranks risk well. Translate the gap into exposure and expected loss using separate assumptions for recovery and exposure at default. Investigate both approval decisions and subsequent portfolio performance; a model that approves more high-risk borrowers can change the data used to assess itself. Material drift calls for controlled changes and renewed validation.

Worked example: For a new segment with 10 million of comparable exposure, observed one-year default frequency rises from 2% to 5%. If exposure at default stays at 10 million, expected defaulted exposure rises by (0.05 − 0.02) × 10 million = 300,000. This is not yet an expected credit loss: applying a hypothetical 40% loss given default would give an incremental 120,000, subject to the model's assumptions.

Watch the distinction: A three-percentage-point increase in default frequency does not directly equal credit loss. Recovery, exposure at default, and whether the observed loans are comparable also matter.

Reference: [1]

58. Assess digital-asset redemption risk

A token promising redemption at par creates a cash obligation when holders request redemption. Assess reserve assets by when they can produce cash, not only by their reported value. Cash meets immediate requests; securities may need sale or maturity, and a rushed sale can realize less than their mark. First compare redemption requests with immediately available cash, then identify the amount and timing of additional funding. Next stress the price and sale capacity of reserves under simultaneous redemptions. If holders doubt timely payment at par, further requests may accelerate, making a liquidity problem more severe. Legal rights, reserve custody, and operational settlement can also affect access to funds. Separate a temporary cash shortfall from a true asset-value shortfall; either can undermine confidence.

Worked example: A hypothetical issuer receives 100 million of same-day redemption requests and has 70 million of immediately usable cash. Its immediate cash gap is 100 − 70 = 30 million. If it sells reserves marked at 32 million for 30 million cash today, the requests can be met, but the sale realizes a 2 million loss relative to that mark. The remaining reserve position must then be reassessed.

Watch the distinction: Reserves valued at or above outstanding tokens do not guarantee same-day redemption. Their cash conversion time and executable sale price determine immediate payment capacity.

Reference: [1]

59. Examine private-credit valuation lag

Private-credit loans often trade infrequently, so their reported values may update more slowly than prices of comparable traded debt. A stable mark can reflect limited observations rather than stable economic value. When borrower conditions change, examine cash-flow forecasts, covenant status, collateral, refinancing needs, and the discount rate or spread used in valuation. A traded comparison provides a diagnostic signal, but differences in seniority, security, maturity, and borrower quality must be adjusted before treating its price as a direct valuation. Separate measurement lag from genuine illiquidity: a loan may be worth less even if no sale occurs. For portfolio risk, infrequent marks can also suppress measured volatility and correlation. Challenge unchanged valuations with current borrower evidence and plausible exit prices.

Worked example: A private loan remains marked at 100 per 100 of face value, while a broadly comparable traded loan falls from 100 to 92. The visible difference is 100 − 92 = 8 price points, or 8% of face value. That is a reason to review the private loan's cash flows and spread. It is not automatically an 8-point write-down because the loans may differ.

Watch the distinction: An unchanged private mark does not prove that market or credit risk is low. Slow price discovery can conceal a change until valuation assumptions are updated.

Reference: [1]

60. Trace nonbank liquidity feedback

Leverage can turn an initial market decline into a broader liquidity shock. Falling asset prices reduce collateral value and trigger margin calls. A fund needing cash may sell assets quickly; its sales can push prices lower, reducing collateral values for other investors and generating more calls. Trace the loop in order: price move, collateral or margin demand, cash need, asset sale, price impact, then new demands. Check whether sales settle soon enough and how much cash remains after meeting the call. Selling a marked position below its mark crystallizes a loss and may weaken the fund's capital, even if the immediate call is paid. Risk analysis should consider common holdings across funds and the possibility that many sellers act at once.

Worked example: A fund owes a 15 million margin call today. It sells bonds marked at 20 million but receives 18 million after price impact, assuming immediate settlement and no fees. It can pay the 15 million call and has 3 million cash left. Relative to the 20 million mark, the sale realizes a 2 million loss. Lower bond prices could also reduce peers' collateral values.

Watch the distinction: Covering today's margin call does not end the risk. The sale can realize a loss, weaken remaining collateral, and contribute to further calls across shared holdings.

Reference: [1]

Official references and further reading

FRM® Exam Information, Steps to Earn Certification | GARP

All study guides

Choose your exam version

FRM Part I

FRM Part II

Foundations of Risk Management

1. Identify the source of risk

2. Turn appetite into limits

3. Assign risk decisions clearly

4. Recognize diversification limits

5. Trace systemic spillovers

6. Compare return with risk capital

Quantitative Analysis

7. Update a probability with evidence

8. Calculate expected loss variability

9. Build portfolio variance

10. Read a distribution's tail

11. Interpret a regression slope

12. Update volatility over time

Financial Markets and Products

13. Value a forward payoff

14. Follow futures margin

15. Read a swap's cash flows

16. Separate option payoff from profit

17. Allocate securitization losses

18. Measure a repo haircut

Valuation and Risk Models

19. Discount promised bond cash flows

20. Approximate rate-driven price changes

21. Price a one-step option

22. Hedge an option's first-order exposure

23. Estimate value at risk from scenarios

24. Measure losses beyond VaR

Market Risk Measurement and Management

25. Map positions to market factors

26. Reprice under historical moves

27. Use sensitivity and stress together

28. Investigate VaR exceptions

29. Match interest-rate sensitivity

30. Recognize basis risk

Credit Risk Measurement and Management

31. Calculate expected credit loss

32. Spot credit concentration

33. Interpret default-model mechanisms

34. Value a credit rating migration

35. Link exposure to CVA

36. Account for collateral and wrong-way risk

Operational Risk and Resilience

37. Classify an operational loss

38. Assess controls against a scenario

39. Choose an actionable risk indicator

40. Combine loss data with scenarios

41. Plan for service disruption

42. Challenge model use

Liquidity and Treasury Risk Measurement and Management

43. Separate funding and market liquidity

44. Construct a cash-flow ladder

45. Estimate buffer survival

46. Stress funding assumptions

47. Price funding into business decisions

48. Sequence contingency funding

Risk Management and Investment Management

49. Relate beta to required return

50. Measure active-return variability

51. Allocate portfolio risk

52. Find hidden factor exposures

53. Calculate drawdown

54. Attribute active performance

Current Issues in Financial Markets

55. Translate climate scenarios into exposure

56. Map shared technology dependencies

57. Monitor AI-driven model change

58. Assess digital-asset redemption risk

59. Examine private-credit valuation lag

60. Trace nonbank liquidity feedback

Official references and further reading

Frequently Asked Questions

Related Study Guides

Certified Government Financial Manager (CGFM)

Professional Risk Manager (PRM)

Associate in Risk Management (ARM)