Why Nonlinear Effects Matter in Stress Testing
A stress scenario can change a balance sheet by more than the sum of its individual shocks. This happens when the calculation includes a threshold, a contractual payoff that changes shape, a rating step, or a limit on what can be liquidated. Before a trigger is reached, a small additional shock may have little effect; just beyond it, the result can change sharply.
Earlier material on scenario drivers, cross-risk contagion, liquidity projections, and capital resources provides useful context for constructing an integrated stress. Here the focus is narrower: how to calculate and interpret nonlinear effects once the stressed inputs have been specified. The calculations below are deterministic scenario projections, not probability forecasts. Unless an example explicitly states otherwise, amounts are in millions of dollars; there is no discounting, interest-rate assumption, mortality assumption, or probability distribution in these one-period calculations.
A practical way to test for nonlinearity is to write the calculation as a function of its inputs before inserting stressed values. Look for maximum or minimum operators, conditional rules, piecewise rates, caps, floors, and capacity limits. Then apply the complete combined scenario to the function. Calculating each shock separately and adding the results may not reproduce the scenario outcome.
Option-Like Liabilities and Floor Effects
A guarantee that makes up a shortfall below a stated floor has an option-like payoff. If the reference asset value after stress is \(A\) and the guaranteed floor is \(K\), the payment is the positive part of the shortfall. For a one-period deterministic stress, the liability is \(L=\max(0,K-A)\). This is a simplified payoff calculation: it assumes the guarantee is fully enforceable, ignores discounting and expenses, and treats the stressed asset value and floor as known.
The piecewise definition explains why a linear approximation can fail. When \(A>K\), a small reduction in \(A\) may still leave the liability at zero. Once \(A\) falls below \(K\), each additional dollar of decline increases the liability by one dollar under this simplified payoff. The change in slope at \(A=K\) is the kink in the payoff.
Worked Example: A Guarantee Crosses Its Floor
A hypothetical insurer has a one-period guarantee with a floor of $100 million. The reference assets are worth $120 million before stress. A combined scenario applies a 10% decline first and then a further 8% decline to the remaining value. Calculate the stressed liability and compare it with the liability before stress. The scenario is deterministic, with no discounting, mortality, or probability distribution.
Method. Apply the sequential percentage shocks to the asset value, then evaluate the full payoff function. The mistake this avoids is adding the percentage declines as though both were applied to the original value.
After the first shock, the asset value is \(120(1-0.10)=120(0.90)=108\) million. The second decline applies to $108 million, so the final asset value is \(108(1-0.08)=108(0.92)=99.36\) million. Equivalently, \(120(0.90)(0.92)=99.36\) million.
The stressed liability is \(L=\max(0,100-99.36)=\max(0,0.64)=0.64\) million. Before stress, \(L=\max(0,100-120)=0\), so the scenario increases the liability by $0.64 million.
Check. The final asset value is $0.64 million below the $100 million floor, exactly matching the calculated guarantee payment. Adding the declines to get 18% would incorrectly produce an asset value of \(120(0.82)=98.4\) million and a liability of $1.60 million.
Collateral Triggers: A Threshold Becomes a Cash Requirement
Collateral arrangements can create an abrupt liquidity need. For example, a contract may require a collateral call when the ratio of eligible collateral to exposure falls below a specified threshold. If the call restores the ratio to a target \(r\), eligible collateral is \(C\), and exposure is \(E\), the required additional collateral is zero when \(C/E\ge r\). When \(C/E<r\), the required amount is \(rE-C\), assuming the collateral is measured on the same basis and there are no other contractual adjustments.
This formula describes the amount needed to restore the target, not necessarily an economic loss. A call can encumber cash or other assets and reduce immediately available liquidity even if the collateral remains an asset of the posting institution. The contractual trigger and the definition of eligible collateral must be checked before applying the formula.
Worked Example: A Collateral Ratio Falls Below Its Trigger
In a hypothetical bilateral contract, eligible collateral is initially $132 million and exposure is $100 million. A stress reduces eligible collateral by 15% and increases exposure by 10%. If the stressed collateral ratio is below 1.10, a call restores it to 1.10. Calculate the call. Assume a deterministic one-period scenario, with no discounting, interest, mortality, or probability distribution.
Method. Recalculate both sides of the ratio under stress, test the contractual threshold, and then calculate the amount needed to restore the target. This avoids checking only the collateral decline while overlooking the simultaneous exposure increase.
Stressed collateral is \(132(1-0.15)=132(0.85)=112.2\) million. Stressed exposure is \(100(1+0.10)=110\) million. The stressed ratio is \(112.2/110=1.02\), which is below the trigger of 1.10.
Collateral at the target ratio would be \(1.10(110)=121\) million. The call is therefore \(\max(0,121-112.2)=8.8\) million.
Check. After posting $8.8 million, eligible collateral is \(112.2+8.8=121\) million, and \(121/110=1.10\), the contractual target. Posting $8.8 million meets the target exactly; it should not automatically be described as an $8.8 million loss.
Rating Downgrades and Step Changes in Capital
A rating-based capital rule may assign different charge rates to different rating categories. If a stressed indicator crosses a downgrade threshold, the applicable rate can change abruptly. This is distinct from a gradual increase in exposure: the rating rule itself changes the calculation. A simplified charge is the stressed exposure multiplied by the charge rate for the rating that applies after stress.
Worked Example: A Combined Stress Triggers a Higher Rating Charge
A hypothetical portfolio has a base rating indicator of 3.60. A catastrophe scenario adds 0.25 indicator points, and a market scenario adds 0.20 points. A downgrade occurs if the combined indicator is greater than 4.00. Capital is charged at 3% of exposure before downgrade and 5% after downgrade. Base exposure is $250 million and rises by 8% in the combined stress. Calculate stressed capital and separate the exposure effect from the downgrade effect. This is a deterministic rule calculation; there is no discounting, mortality, or probability distribution.
Method. Aggregate the indicator stresses, test the downgrade condition, and apply the rate to stressed exposure. To isolate the step effect, also calculate the charge on stressed exposure at the original rating rate. This avoids applying the higher rate to base exposure or treating the rating change as a gradual rate increase.
The combined indicator is \(3.60+0.25+0.20=4.05\). Since \(4.05>4.00\), the downgrade condition is met. Stressed exposure is \(250(1.08)=270\) million.
At the downgraded 5% rate, stressed capital is \(270(0.05)=13.50\) million. At the original 3% rate on the same stressed exposure, it would be \(270(0.03)=8.10\) million. Thus the exposure increase, at the original rate, raises capital from \(250(0.03)=7.50\) million to $8.10 million, an increase of $0.60 million. The rating step adds \(13.50-8.10=5.40\) million. Relative to the original $7.50 million charge, the total increase is \(13.50-7.50=6.00\) million.
Check. The $0.60 million exposure effect plus the $5.40 million rating-step effect equals the $6.00 million total increase. The decomposition uses the original rate for the exposure effect and then applies the rate step to stressed exposure; a different ordering could produce a different allocation between the two effects, but the total remains $6.00 million.
Deriving the General Threshold Calculation
The collateral-call calculation is a useful general pattern. Let \(g\) be the shortfall from a required target. If there is no shortfall, the required action is zero; if there is a shortfall, the required action equals that amount. These two cases are exactly the definition of the positive-part function.
For a collateral target \(r\), the target amount is \(rE\). The gap between target and current eligible collateral is \(g=rE-C\).
If \(g\le 0\), current collateral already meets the target and the call is zero. If \(g>0\), the call must equal \(g\) to reach the target.
The call is \(\max(0,g)=\max(0,rE-C)\). This follows directly from the two contractual cases, rather than from an assumption that calls rise smoothly with every input.
The same reasoning applies to guarantees, deductibles, minimum capital requirements, and limits: first define the relevant gap, then determine which side of the threshold applies. For a combined scenario, the inputs must be stressed together before the condition is tested. A threshold should not be applied to an unstressed input simply because that input is easier to observe.
Liquidation Constraints and Nonlinear Liquidity Results
Asset liquidation may be limited by trading capacity, settlement timing, or market depth. A haircut also means that each dollar of market value sold produces less than one dollar of cash. If asset \(j\) has market value \(V_j\), maximum permitted sale \(q_j^{\max}\), and haircut \(h_j\), then selling market value \(q_j\), where \(0\le q_j\le q_j^{\max}\), produces \((1-h_j)q_j\) in cash. The capacity makes the cash proceeds a constrained, piecewise function: selling more than the permitted amount is not an available way to meet the need.
Worked Example: A Cash Need Exceeds Constrained Liquidation Proceeds
A hypothetical institution needs $45 million in cash under stress. It can sell up to $20 million of Asset A at a 2% haircut and up to $25 million of Asset B at a 10% haircut. Assume the stated capacities are maximum market values that can be sold during the stress horizon, and that both can be fully used. Calculate the maximum cash proceeds and any shortfall. This deterministic calculation has no discounting, interest, mortality, or probability distribution.
Method. Apply each haircut to the permitted sale amount, then compare total cash proceeds with the required amount. This avoids treating market value sold as cash received or assuming that unconstrained assets can be liquidated immediately.
Asset A produces \(20(1-0.02)=20(0.98)=19.60\) million in cash. Asset B produces \(25(1-0.10)=25(0.90)=22.50\) million. Maximum total proceeds are \(19.60+22.50=42.10\) million.
The cash shortfall is \(45-42.10=2.90\) million. Although the two assets have a combined market value of $45 million at their stated sale capacities, their proceeds are only $42.10 million after haircuts.
Check. The total haircut cost on the permitted sales is \(20(0.02)+25(0.10)=0.40+2.50=2.90\) million. Thus $45 million of market value sold less $2.90 million of haircuts gives $42.10 million of cash, consistent with the proceeds calculation. Ignoring capacities and haircuts would wrongly conclude that the cash need can be met.
Common Mistakes and Interpretation
- Adding sequential percentage changes. A 10% decline followed by an 8% decline gives a multiplier of \(0.90(0.92)=0.828\), not \(1-0.10-0.08=0.82\). The order and base of each shock matter.
- Applying a trigger before stressing all its inputs. A collateral ratio depends on both collateral and exposure. Stressing only the numerator can miss a trigger caused by an increase in the denominator.
- Using the wrong inequality at the boundary. Whether a rule activates at “greater than” or “greater than or equal to” the threshold is contractual. The exact boundary condition must be retained.
- Calling every cash requirement a loss. A collateral call can encumber an asset without reducing net worth by the same amount. Report liquidity use and economic loss as distinct quantities unless the scenario assumptions establish that they are equal.
- Ignoring constraints in a liquidity projection. A portfolio may have sufficient aggregate market value but insufficient cash proceeds within the stress horizon because of sale limits, haircuts, or settlement constraints.
- Extrapolating a local sensitivity across a kink. A sensitivity estimated before a guarantee floor is crossed or before a rating change occurs may not describe the result after the trigger.
Check Your Understanding
For each hypothetical deterministic scenario, apply the stated rule after stressing the inputs. Amounts are in millions of dollars; no discounting, interest, mortality, or probability distribution is assumed unless stated.
- A guarantee pays the amount by which $120 exceeds the stressed asset value, if positive. Assets start at $150 and fall 12%, then fall another 10% on the remaining value. What is the payment? (Answer: $1.20 million)
- Eligible collateral is $90 and exposure is $70. Stress reduces collateral by 10% and increases exposure by 20%. A call is made if the ratio is below 1.05 and restores the ratio to 1.05. Calculate the call. (Answer: $7.20 million)
- A rating indicator starts at 2.70. Two scenario components add 0.60 and 0.80 points. Downgrade occurs above 4.00. Exposure starts at $160 and rises 5%; the charge rate changes from 2.5% to 4.5% upon downgrade. What is the stressed charge, and how much of it is the rate-step increment relative to applying 2.5% to stressed exposure? (Answer: $7.56 million stressed charge; $3.36 million rate-step increment)
- A stressed cash need is $30. Asset A permits a maximum market-value sale of $12 at a 2% haircut; Asset B permits $18 at an 8% haircut. What is maximum cash proceeds and the resulting shortfall? (Answer: $28.32 million proceeds; $1.68 million shortfall)
- A rule requires an additional payment only when an indicator is strictly greater than 5.00. Under a scenario the indicator is exactly 5.00. Does the payment activate? Explain why the boundary matters. (Answer: No; “strictly greater than” excludes equality, so the threshold payment is not activated.)