Model Risk in Hedge Design and Stress Results

Why Model Risk Matters in Hedge Design

A hedge ratio is only as reliable as the valuations and sensitivities used to calculate it. A change in discounting, cash-flow projection, volatility assumption, or mapping of market factors can change the estimated exposure and the estimated hedge payoff. The resulting hedge may look precise while depending on a narrow set of assumptions. Stress results can be just as sensitive: a scenario can appear well hedged under one valuation model and poorly hedged under another.

The issue is not that a model must reproduce every future market outcome. It is that decisions should not be presented as robust when modest, plausible changes in assumptions materially alter the hedge size or the sign and scale of a stress loss. Model risk analysis makes those dependencies visible. It complements the earlier material on market stress scenarios, correlated shocks, nonlinear positions, and extreme-move mitigation; it does not replace judgment about which scenarios are relevant.

Definition: Model risk in hedge design is the risk that errors, limitations, or assumptions in the models used to value an exposure, value a hedge, estimate sensitivities, or construct scenarios lead to an inappropriate hedge decision or a misleading assessment of residual risk. Parameter uncertainty is uncertainty about model inputs, such as volatility or a sensitivity estimate. Structural uncertainty concerns the model form or representation itself, such as whether important nonlinear effects or market factors are omitted.

For consistent calculations, define $L_s$ as the exposure's loss in scenario $s$, and $G_s$ as the hedge's gain per unit in that same scenario. A hedge position of $h$ units produces net loss $N_s=L_s-hG_s$. A positive $N_s$ is a net loss; a negative $N_s$ is a net gain. This sign convention assumes the hedge is held in the direction that gains when $G_s$ is positive. It is essential to keep the exposure loss and hedge gain on the same horizon, in the same currency, and under the same scenario definition.

Hedge Ratios Depend on the Valuation Model

For a single specified shock, if the modeled exposure loss is $L>0$ and one unit of the hedge is modeled to gain $G>0$, the ratio that offsets that loss is $h=L/G$. This is a local or scenario-specific calculation, not a universal property of the exposure. It presumes the loss and gain have been valued consistently and that the hedge response scales linearly with the number of units over the position range being considered.

N=L-hG,\qquad N=0 \ \Longrightarrow\ h=\frac{L}{G},\quad G\ne 0.

A common source of instability is that $L$ and $G$ may both depend on modeling choices. A different discount curve or projection of cash flows can change the exposure's value response. A different volatility input or treatment of optionality can change the hedge instrument's value response. Even if both valuations are reasonable, their ratio can change materially. The ratio should therefore be reported with its valuation basis and shock, not as a context-free number.

Worked Example: Valuation Assumptions Change the Hedge Ratio

A hypothetical insurer estimates that a specified parallel rate shock produces an exposure loss of $480,000 per basis point. Under the same shock and valuation date, one futures position is estimated to gain $40,000 per basis point. Calculate the offsetting hedge ratio. Then repeat the calculation under an alternative valuation basis that estimates an exposure loss of $440,000 and a hedge gain of $50,000 per basis point. Assume linear responses over the stated shock range, the same currency and horizon, and no separate cash flows.

Method and base valuation. Use $h=L/G$, since the target is zero net loss for this one shock. The units are consistent: both sensitivities are dollars per basis point.

h_{\text{base}}=\frac{\$480{,}000\text{ per bp}}{\$40{,}000\text{ per bp per position}}=12\text{ positions}.

For a 50-basis-point shock, the modeled exposure loss is $50\times \$480{,}000=\$24{,}000{,}000$. The gain from 12 positions is $12\times 50\times \$40{,}000=\$24{,}000{,}000$. The resulting net loss is $24-24=\$0$ million under this linear model.

Alternative valuation basis. The ratio is now $h_{\text{alt}}=\$440{,}000/\$50{,}000=8.8$ positions. At 50 basis points, the exposure loss is $50\times \$440{,}000=\$22{,}000{,}000$; the gain from 8.8 positions is $8.8\times50\times\$50{,}000=\$22{,}000{,}000$. The net loss is again zero within the alternative model.

The two internally offsetting calculations recommend 12 and 8.8 positions. The difference is not a rounding issue; it is evidence that hedge sizing depends on the valuation basis. Common mistake avoided: applying the base-model ratio to the alternative model and then claiming it is still an exact offset. With 12 positions under the alternative basis, the gain would be $12\times50\times\$50{,}000=\$30{,}000{,}000$, giving a net gain of $8,000,000 rather than a zero loss.

Comparing Hedge Performance Across Scenarios

A hedge calibrated to one shock need not protect against a different shock. To compare outcomes, calculate the exposure loss and hedge gain separately for each scenario, then apply the same position $h$. A model that reports only the calibrated scenario can conceal poor performance under a different combination of market-factor movements.

Worked Example: Checking a Fixed Hedge Across Three Outcomes

A hypothetical portfolio has exposure losses of $1,000, $3,000, and $5,000 in three modeled outcomes. One unit of a proposed hedge has gains of $1,000, $2,000, and $3,000 in those outcomes. The desk proposes a fixed position of $h=2$ units. Calculate each net loss and explain what the results do and do not establish. Assume all amounts are on the same horizon and ignore costs and cash flows outside the stated outcomes.

Method. For each outcome, use $N_s=L_s-hG_s$. In thousands of dollars, the calculations are:

N_1=1-2(1)=-1,\qquad N_2=3-2(2)=-1,\qquad N_3=5-2(3)=-1.

Thus the net losses are $-\$1{,}000$, $-\$1{,}000$, and $-\$1{,}000$: the modeled net result is a gain of $1,000 in each outcome. The check is direct: the hedge gains are $2,000, $4,000, and $6,000, each $1,000 greater than the corresponding exposure loss.

This conclusion applies only to the stated outcome values and hedge position. For example, suppose a revised valuation model estimates the hedge gains as $800, $1,600, and $2,400 instead, while exposure losses are unchanged. At two units, the revised net losses are $1-2(0.8)=-0.6$, $3-2(1.6)=-0.2$, and $5-2(2.4)=0.2$ thousand dollars. The third outcome is now a net loss of $200. Common mistake avoided: reporting positive $1,000 losses by subtracting the exposure loss from the hedge gain, or by ignoring the convention $N=L-hG$. A negative result here is a net gain, not a loss.

Fitting a Hedge to Several Scenarios

When no single scenario represents the full objective, one possible method is to choose $h$ to minimize weighted squared net losses across a selected set of scenarios. Let $w_s$ be nonnegative scenario weights and suppose the same hedge unit has gain $G_s$ in scenario $s$. The objective is a least-squares fit:

Q(h)=\sum_s w_s(L_s-hG_s)^2.

If $\sum_s w_sG_s^2>0$, differentiating with respect to $h$ and setting the derivative equal to zero gives the minimizing position:

\frac{dQ}{dh}=-2\sum_s w_sG_s(L_s-hG_s)=0 \quad\Longrightarrow\quad h^*=\frac{\sum_s w_sL_sG_s}{\sum_s w_sG_s^2}.

This derivation shows why the formula is a fit to the chosen scenarios: it makes the weighted residuals $L_s-hG_s$ orthogonal, in the weighted calculation, to the hedge gains $G_s$. It does not guarantee that every scenario has a small net loss, that the worst loss is minimized, or that the weights are appropriate. A squared-loss objective gives large residuals more influence because they are squared, but it is not itself a tail-risk measure. The selected scenarios, weights, and objective should be stated rather than treated as neutral facts.

Worked Example: A Weighted Scenario Hedge Fit

A hypothetical insurer considers two scenarios. The exposure losses, in thousands of dollars, are $L_1=2$ and $L_2=4$; the hedge gains per unit are $G_1=1$ and $G_2=3$. Give the scenarios equal weights and choose $h$ to minimize the sum of squared net losses. Then calculate both residual net losses. Assume the scenario valuations are on the same horizon and use the same currency.

Method. Apply the least-squares formula with $w_1=w_2=1$:

h^*=\frac{(2)(1)+(4)(3)}{1^2+3^2} =\frac{2+12}{1+9} =\frac{14}{10}=1.4\text{ units}.

The first net loss is $N_1=2-1.4(1)=0.6$ thousand dollars, or $600. The second is $N_2=4-1.4(3)=4-4.2=-0.2$ thousand dollars, or a net gain of $200. The squared-loss objective is $0.6^2+(-0.2)^2=0.36+0.04=0.40$ in squared thousands of dollars.

As a check on the minimization, the derivative condition is $N_1G_1+N_2G_2=0.6(1)+(-0.2)(3)=0$, as required. Common mistake avoided: dividing the sum of exposure losses by the sum of hedge gains, which would give $6/4=1.5$. That ratio does not minimize the stated squared-loss objective; the correct calculation uses the products $L_sG_s$ and squares $G_s^2$.

Stress Shocks Can Reveal Basis and Model Dependence

A parallel shock and a curve twist are different scenarios, not interchangeable versions of the same test. A hedge may offset a parallel movement while leaving substantial residual risk when market factors move in different directions. This can arise from basis risk, different factor exposures, or an incomplete representation of the hedge response. Scenario sensitivity is therefore a check on the design, not merely a second presentation of the original calculation.

Worked Example: A Parallel Hedge That Does Not Generalize to a Twist

For a hypothetical portfolio, let $x$ and $y$ denote shocks to two market factors. Under a specified linear valuation approximation, the exposure loss is $L=10x+6y$ thousand dollars, and each hedge unit gains $G=4x+2y$ thousand dollars. First find the position that offsets a parallel shock $(x,y)=(10,10)$. Then calculate the net result under a twist $(20,-10)$ using that same position. Assume these formulas apply to both stated scenarios and that positive $G$ is a hedge gain.

Parallel shock. The exposure loss is $10(10)+6(10)=100+60=160$ thousand dollars. The gain per hedge unit is $4(10)+2(10)=40+20=60$ thousand dollars. Therefore the offsetting position is $h=160/60=8/3$ units.

Twist. The exposure loss is $10(20)+6(-10)=200-60=140$ thousand dollars. The gain per unit is $4(20)+2(-10)=80-20=60$ thousand dollars. The hedge gain from $8/3$ units is $(8/3)(60)=160$ thousand dollars. Hence $N=140-160=-20$ thousand dollars, a net gain of $20,000.

The arithmetic is internally consistent: the position exactly offsets the parallel shock and produces a net gain under this particular twist. That favorable result does not prove the hedge is robust; another shock direction or revised sensitivity estimates could reverse the residual. Common mistake avoided: using two units as the parallel hedge. The parallel loss is $160,000 and the gain per unit is $60,000, so the ratio is $160/60=8/3$, not 2. Under the twist, the calculated net result is a $20,000 gain, not a $20,000 loss.

Practical Sensitivity Checks

A useful model-risk review changes one assumption at a time where possible, then tests selected combinations where interactions may matter. The purpose is to identify which assumptions drive the decision and whether the direction and scale of the results remain credible. A check should retain a clear baseline, state what changed, and compare both the recommended hedge and the residual stress loss.

Fix the measurement basis.
Record the valuation date, horizon, currency, market factors, and sign convention for exposure losses and hedge gains. A comparison is not meaningful if the quantities are measured on inconsistent bases.

Vary important valuation inputs.
Recalculate exposure and hedge values under plausible alternatives to discounting, cash-flow projection, volatility, or other material assumptions. Track how the hedge ratio $L/G$ changes.

Change the shock shape.
Test more than the shock used to calibrate the hedge. Where relevant, compare parallel movements with twists, basis movements, or combinations of factors, using coherent scenario definitions.

Reconcile the P&L.
For each case, show $L_s$, $G_s$, $hG_s$, and $N_s=L_s-hG_s$. Check units, signs, and whether the loss and gain are valued over the same horizon.

Explain decision sensitivity.
Identify whether alternative assumptions change the hedge size, leave a material residual, or change a net loss into a gain. Escalate conclusions that depend strongly on one uncertain assumption rather than presenting a single estimate as definitive.

Sensitivity checks are not a substitute for selecting a sound model or a relevant stress. They are a disciplined way to expose dependence on assumptions and to distinguish a stable conclusion from one that is fragile. A zero modeled loss under one shock is a conditional result, not proof that all material market risk has been removed.

Key takeaway: A hedge ratio is conditional on the valuation model, shock, horizon, and objective used to calculate it. Report those conditions, test plausible alternatives, and verify every stress result with $N_s=L_s-hG_s$. A materially changing ratio or residual is evidence to investigate, not a detail to hide through rounding.

Check Your Understanding

Use the stated assumptions and keep losses and hedge gains on the same basis in each calculation.

A model estimates exposure loss of $270,000 per basis point and hedge gain of $45,000 per basis point per position. Find the offsetting position. Under an alternative model the corresponding sensitivities are $330,000 and $40,000. What are the two ratios? (Answer: 6 positions under the first model; 8.25 under the alternative.)
In three outcomes, exposure losses in thousands are $1,3,5$, and hedge gains per unit are $1,2,3$. A fixed hedge has 2 units. Calculate the three net losses using $N=L-hG$. (Answer: $-1,-1,-1$ thousand dollars; a net gain of $1,000 in each outcome.)
Exposure loss is $L=8x+5y$ thousand dollars and hedge gain per unit is $G=3x+y$ thousand dollars. Find the hedge ratio for $(x,y)=(10,10)$, then find the net loss under $(15,-5)$ using that ratio. (Answer: $h=13/4=3.25$ units; twist net loss $-35$ thousand dollars, a net gain of $35,000.)
Two equally weighted scenarios have exposure losses $L_1=3,L_2=6$ and hedge gains per unit $G_1=1,G_2=2$, all in thousands of dollars. Find the least-squares hedge ratio and verify the residuals. (Answer: $h=3$ units; net losses $0$ and $0$ thousand dollars.)
An exposure loss and hedge gain per unit are both estimated at $900,000 for a particular shock. A revised valuation estimates the exposure loss at $810,000 and hedge gain at $720,000. Find the two offsetting ratios and state one limitation of the comparison. (Answer: 1 position under the first valuation and 1.125 under the revised valuation; each ratio offsets only its stated shock and valuation assumptions.)

What You'll Learn