Integrated Market Risk Measurement Problems

Why Integrated Market Risk Measurement Matters

An insurer’s market risk rarely comes from one factor in isolation. A portfolio may hold bonds exposed to both interest rates and credit spreads, equities exposed to market returns, and foreign assets whose domestic value changes with exchange rates. A useful risk result must put these exposures on a common footing, preserve the sign of each gain or loss, and make clear whether it is a sensitivity estimate, a scenario result, or a probability-based measure.

This tutorial integrates those calculations rather than treating each exposure as a separate result. The methods are aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus. All portfolios and figures below are hypothetical.

Definition: Let $\Delta V$ denote the change in portfolio market value or insurer surplus over a specified horizon. Define the corresponding loss as $L=-\Delta V$, so a gain has negative loss and a decline in value has positive loss. A sensitivity estimate approximates $\Delta V$ using factor exposures and factor changes. Scenario repricing recalculates values under a specified joint market scenario. VaR at confidence level $q$ is the $q$-quantile of the loss distribution: $\operatorname{VaR}_q(L)=\inf\{l:P(L\leq l)\geq q\}$.

Combine Factor Sensitivities With Consistent Signs

A sensitivity calculation is a local approximation. It is useful when factor changes are sufficiently small and the portfolio’s response is approximately linear over the changes being considered. The calculation should specify the direction of every market move and the sign convention for every exposure. For example, a positive net foreign-currency position gains when that currency appreciates in domestic-currency terms, while a bond generally loses value when its yield or credit spread rises.

The factor contributions are added to estimate the total change in value. The sign of the result is essential: sum the signed changes first, then reverse the sign to report loss. A sensitivity estimate is not the same as a fully repriced scenario, particularly for large moves or nonlinear positions.

Worked Example: Combining Four First-Order Exposures

A hypothetical insurer estimates the following one-day exposures. Its net rate DV01 is $6,000 per basis point, so a one-basis-point rate increase reduces value by $6,000. Its beta-adjusted equity exposure is $2.50 million. Its net foreign-currency position is worth $3.00 million in domestic currency. Its spread-sensitive bond portfolio is worth $5.00 million and has spread duration 3. Under a joint scenario, rates rise 35 basis points, equities fall 6%, the foreign currency appreciates 4% against the domestic currency, and credit spreads widen 50 basis points. Estimate the change in value and the loss using first-order sensitivities.

Method and assumptions. Use linear sensitivities, apply the scenario moves to the stated exposures, and assume no interaction or nonlinear terms. Treat the FX percentage as the change in domestic value of the foreign-currency position.

Rates. The rate increase is 35 basis points, so the value change is $-\$6{,}000\times35=-\$210{,}000$.

Equities. A 6% decline on the $2,500,000 beta-adjusted exposure gives $\$2{,}500{,}000\times(-0.06)=-\$150{,}000$.

Foreign exchange. The foreign currency appreciates, so the long position gains: $\$3{,}000{,}000\times0.04=+\$120{,}000$.

Credit spreads. For a small spread change, the first-order bond value change is approximately negative spread duration times market value times the spread change. The 50-basis-point widening is 0.005, giving $-3\times\$5{,}000{,}000\times0.005=-\$75{,}000$.

Aggregate. The signed value change is $-\$210{,}000-\$150{,}000+\$120{,}000-\$75{,}000=-\$315{,}000$. Therefore the estimated loss is $L=-\Delta V=+\$315{,}000$.

Check and common mistake. The three adverse contributions total $435,000 and the FX gain offsets $120,000, leaving a $315,000 loss. Treating the FX appreciation as a loss would incorrectly report $555,000. This estimate is local and does not include convexity or interactions.

Scenario Repricing and Reconciliation

Scenario repricing applies one joint set of market-factor changes to the valuation model and compares the resulting portfolio value or surplus with its starting value. If the portfolio includes nonlinear positions, repricing can capture effects that a first-order sensitivity calculation misses. The factor contributions in an attribution may depend on the model’s allocation convention; when interaction effects are present, they should be allocated consistently so the contributions reconcile to the total change.

The basic check is direct: the sum of signed contributions must equal the change in value being explained. If the starting and ending values are known, calculate the loss from their difference independently. That check can reveal a sign error, an omitted factor, or an attribution that does not reconcile.

Worked Example: Checking a Joint Repricing Result

A hypothetical insurer begins with surplus of $8,400,000. A joint scenario repricing changes rates, equity values, exchange rates, and credit spreads. The valuation model reports the following sequentially attributed changes, in thousands of dollars: rates, $-420.125000$; equities, $-310.250000$; FX, $-145.000000$; and spreads, $-112.000390$. The attribution includes interaction effects according to the model’s stated sequential convention. Find the total loss and ending surplus.

Method. Sum the signed, model-reported contributions, reverse the sign to obtain the loss, and independently subtract that loss from starting surplus. Because the contributions are in thousands, convert the final result to dollars before rounding to cents.

The total change, in thousands, is

\Delta V=-420.125000-310.250000-145.000000-112.000390=-987.375390.

Thus the loss is $987.375390$ thousand dollars. Converting without first rounding the thousand-dollar figure gives $987.375390\times1{,}000=\$987{,}375.390$, which rounds to $987,375.39 to the nearest cent.

The ending surplus is $\$8{,}400{,}000.00-\$987{,}375.39=\$7{,}412{,}624.61$. In thousands, this is $8{,}400.000000-987.375390=7{,}412.624610$, or $7,412,624.61 to the nearest cent. The two calculations agree.

Check and common mistake. Rounding $987.375390$ thousand to $987.3754$ thousand before converting to dollars would give $987,375.40. That is not the correctly rounded cent amount: keep the given precision through the conversion, then round once at the end. Also, do not add independently estimated factor effects to this attribution; these are already the joint repricing model’s reconciled contributions.

Parametric VaR Across Correlated Factors

A parametric VaR calculation requires an explicit model for the joint factor changes and the portfolio’s response to them. In the linear-normal model used here, the loss is a linear function of normally distributed market-factor shocks. Its mean and variance determine its normal loss distribution. The model is not appropriate without qualification when factor shocks are materially non-normal, the portfolio response is strongly nonlinear, or the chosen horizon and factor distribution do not represent the risk being measured.

Let $Z_i$ be the standardized shock in factor $i$, so each has mean zero and variance one. Let $b_i$ be the portfolio loss, in dollars, associated with a one-standard-deviation increase in that factor, with sign included. If the loss also has expected value $\mu_L$, then $L=\mu_L+\sum_i b_iZ_i$. Write $\Sigma$ for the covariance matrix of the factor shocks.

Theorem: If the factor vector is jointly normal and the portfolio loss is linear in those factors, then $L$ is normal with mean $\mu_L$ and variance $b^\mathsf{T}\Sigma b$. Consequently, $\operatorname{VaR}_q(L)=\mu_L+z_q\sqrt{b^\mathsf{T}\Sigma b}$, where $z_q$ is the $q$-quantile of the standard normal distribution. This formula does not give an exact VaR for a nonlinear portfolio or a non-normal loss distribution.

Why the variance formula holds. Variance of a sum includes both the individual variances and the pairwise covariances. Expanding the variance of $\sum_i b_iZ_i$ therefore gives

\operatorname{Var}\!\left(\sum_i b_iZ_i\right) =\sum_i b_i^2\operatorname{Var}(Z_i) +2\sum_{i<j}b_ib_j\operatorname{Cov}(Z_i,Z_j) =b^\mathsf{T}\Sigma b.

Joint normality and linearity make the resulting loss normal. Standardizing that loss gives $(L-\mu_L)/\sqrt{b^\mathsf{T}\Sigma b}$, which has the standard normal distribution. Taking its $q$-quantile and rearranging gives the stated VaR formula.

Worked Example: Four-Factor Parametric VaR

A hypothetical insurer models one-day loss using four standardized factors: rates, equity returns, FX, and credit spreads. Their respective loss sensitivities per one-standard-deviation shock are $100$, $80$, $-40$, and $60$ thousand dollars. Each factor has variance 1, and every distinct pair has correlation 0.25. Expected one-day loss is $25,000. Assuming jointly normal shocks and a linear portfolio response, calculate 95% VaR.

Method and assumptions. Use the covariance-matrix formula with the signed sensitivity vector $b=(100,80,-40,60)$, measured in thousands of dollars. Use $z_{0.95}=1.644854$. The common pairwise correlation is admissible: the four-factor equicorrelation matrix is positive definite when its correlation is greater than $-1/3$ and less than 1, and 0.25 satisfies those bounds.

The sum of squared sensitivities is $100^2+80^2+(-40)^2+60^2=10{,}000+6{,}400+1{,}600+3{,}600=21{,}600$, in squared thousands of dollars. The sum of the sensitivities is $100+80-40+60=200$. Thus the sum of the six distinct pairwise products is $\bigl(200^2-21{,}600\bigr)/2=9{,}200$.

The variance is the sum of squared sensitivities plus twice the correlation times the sum of pairwise products:

\operatorname{Var}(L)=21{,}600+2(0.25)(9{,}200)=26{,}200 \quad\text{(thousand dollars squared).}

The standard deviation is $\sqrt{26{,}200}=161.86414$ thousand dollars (shown to five decimals for the calculation). Therefore 95% VaR is $25+1.644854(161.86414)=291.242879$ thousand dollars, or $291,242.88 to the nearest cent.

Check and common mistake. The negative FX sensitivity makes its covariance terms with positive sensitivities negative; retaining its sign in the pairwise products is essential. Squaring every sensitivity and then adding only the diagonal terms would omit diversification and concentration effects from correlation. The result is a model-based loss quantile, not a guaranteed maximum loss.

How to Read an Integrated Result

The three calculations answer related but different questions. A sensitivity estimate gives a fast, local response to a specified set of factor moves. Joint scenario repricing gives the model’s value under a particular combined scenario. Parametric VaR gives a loss quantile under a specified probability model. They need not agree: their scenarios, distributional assumptions, treatment of nonlinearity, and horizons may differ.

Before comparing results, check that they use the same portfolio scope, valuation basis, horizon, and sign convention. For a sensitivity or scenario result, verify that gains and losses are signed consistently and that all contributions reconcile. For VaR, verify the factor units, covariance assumptions, and whether the calculation uses mean loss or assumes zero mean. An apparent discrepancy may reflect a genuine difference in method rather than an arithmetic error.

Set the measurement basis.
Identify the portfolio value or surplus, the risk horizon, and whether the requested result is a sensitivity estimate, a scenario loss, or VaR.

Fix signs and units.
State the direction of each factor move and record exposures and shocks in compatible units, such as basis points rather than percentage points for rate changes.

Aggregate and check.
Sum signed changes, reconcile scenario attributions to total repriced value, or include covariance terms in a VaR calculation. Convert and round money only at the end.

Common Integrated-Measurement Traps

Reversing the loss sign. If $\Delta V$ is negative, the portfolio has lost value and $L=-\Delta V$ is positive.
Mixing units. A 50-basis-point spread move is 0.005 in decimal form, not 0.50. Use the spread-duration approximation with a decimal yield change.
Ignoring an exposure’s direction. A long foreign-currency position benefits from appreciation; a short position does not. Preserve signed sensitivities in covariance terms.
Calling a scenario loss VaR. One specified scenario does not supply a probability distribution or a confidence-level quantile.
Rounding too early. Keep available precision through aggregation and currency conversion, and round the reported monetary result only at the final step.

Key takeaway: Integrated market risk measurement depends on consistent signs, units, and assumptions. Sensitivity aggregation, joint scenario repricing, and parametric VaR are complementary methods, but they answer distinct questions and should not be treated as interchangeable.

Check Your Understanding

For each problem, state the requested result with its sign and units.

A portfolio has net rate DV01 of $4,200 per basis point and spread duration 2.5 on $6 million of spread-sensitive bonds. Rates rise 12 basis points and spreads widen 40 basis points. Using first-order sensitivities, what is the total loss? (Answer: $110,400)
An insurer has a long foreign-currency position worth $1.8 million in domestic currency and an equity exposure of $900,000. The currency appreciates 3% and equities fall 5%. Using linear sensitivities, what is the net value change? (Answer: a $9,000 gain)
A joint repricing starts at $2,000,000 surplus and reports signed changes of $-\$80,000$ for rates, $-\$45,000$ for equities, $+\$20,000$ for FX, and $-\$15,000$ for spreads. What is the total loss and ending surplus? (Answer: $120,000 loss; ending surplus $1,880,000)
A linear two-factor normal loss model has expected loss $10,000, signed sensitivities of $120,000 and $-\$50,000$ per standard-deviation shock, and independent unit-variance factors. Using $z_{0.99}=2.3263$, calculate 99% VaR. (Answer: $312,419.00)

What You'll Learn