Stress Testing Portfolios: Historical and Hypothetical Scenarios

Stress testing subjects a portfolio to extreme but plausible market scenarios to assess potential losses beyond what standard risk models capture. While VaR and Expected Shortfall quantify risk under the distribution estimated from recent history, stress tests ask "what if?" -- what if 2008 repeats, what if interest rates spike 300 basis points in a month, what if a major counterparty defaults. Regulators require stress testing for banks and systemically important institutions, but the practice is equally valuable for any quantitative portfolio manager seeking to understand hidden vulnerabilities.

Types of Stress Tests

Historical Scenario Replay

Apply the actual market moves from a historical crisis to the current portfolio. This is the most straightforward approach because the scenarios are drawn from real events, making them credible and interpretable. Key historical scenarios for equity portfolios:

Scenario	Period	S&P 500	10Y UST Yield	VIX	Credit (IG OAS)
Black Monday	Oct 1987	-20.5% (1 day)	-50 bps	+100 pts	+100 bps
LTCM/Russia	Aug-Oct 1998	-19.3%	-80 bps	+25 pts	+150 bps
Dot-Com Peak-Trough	Mar 2000-Oct 2002	-49.1%	-200 bps	+20 pts	+200 bps
GFC	Oct 2007-Mar 2009	-56.8%	-250 bps	+65 pts	+500 bps
COVID Crash	Feb-Mar 2020	-33.9%	-120 bps	+55 pts	+300 bps
2022 Rate Shock	Jan-Oct 2022	-25.4%	+250 bps	+15 pts	+80 bps

Implementation:

Record the daily returns of all risk factors during the historical period

Apply these returns to the current portfolio's factor exposures

Compute the portfolio P&L under each day of the scenario

Report the cumulative loss, maximum drawdown, and recovery time

Critical consideration: The current portfolio may contain assets or exposures that did not exist during the historical period. Synthetic proxies must be constructed for these positions, using factor models or similar assets that did exist.

Hypothetical Scenario Analysis

Construct plausible but unprecedented scenarios that test specific portfolio vulnerabilities. Hypothetical scenarios are particularly valuable for risks that have no historical precedent in recent data. Common hypothetical scenarios: Interest rate shock: Parallel shift of +200 bps across the yield curve over 2 weeks. You can check today's actual Treasury yield curve as your starting point before applying the shock. Estimate impact using duration and convexity: P&L = -Duration delta_y + 0.5 Convexity * (delta_y)^2 For a portfolio with modified duration of 5.5 and convexity of 40: P&L = -5.5 0.02 + 0.5 40 * 0.0004 = -11.0% + 0.8% = -10.2% Credit contagion: Investment-grade spreads widen by 200 bps (from 100 to 300 bps) while high-yield spreads widen by 500 bps. Apply spread duration to credit positions and estimate equity impact through the equity-credit correlation. Liquidity crisis: Model a scenario where bid-ask spreads widen by 10x for illiquid positions, and the portfolio must liquidate 20% of holdings within 5 days. The liquidation cost depends on position size relative to daily volume and the assumed market impact model. Geopolitical shock: Commodity prices spike 40% (oil to $120+), emerging market currencies decline 15-25%, safe-haven assets appreciate, and equity markets decline 15-20%. Specify correlated moves across all risk factors.

Reverse Stress Testing

Instead of asking "what is the loss under scenario X?", reverse stress testing asks "what scenario produces a loss of Y?" This identifies the portfolio's most dangerous vulnerabilities. Methodology:

Define a critical loss threshold (e.g., -20% of portfolio value, or the level that triggers margin calls)

Search over the space of possible scenarios to find the most plausible ones that produce the critical loss

Evaluate whether the identified scenarios are realistic and whether the portfolio should be restructured to avoid them

Optimization formulation: Minimize: ||delta_r - delta_r_historical||^2 (find the closest scenario to normal conditions) Subject to: portfolio_loss(delta_r) >= loss_threshold Where delta_r is the vector of risk factor changes and the objective function ensures the scenario is as close to normal as possible (most plausible) while achieving the required loss. Reverse stress testing often reveals non-obvious vulnerabilities. A portfolio that appears diversified under standard stress tests may be vulnerable to a specific combination of moderate moves (e.g., a simultaneous 10% equity decline, 100 bps rate increase, and 30% oil price spike) that individually seem manageable but collectively produce a critical loss.

Stress Test Implementation Framework

Risk Factor Mapping

Map every portfolio position to underlying risk factors:

Equity positions: Market return, sector returns, style factor returns (value, momentum, quality), individual stock residuals

Fixed income: Interest rate curve (2Y, 5Y, 10Y, 30Y key rates), credit spreads (IG, HY by sector), prepayment rates

FX positions: Individual currency pair returns, volatility surfaces

Commodities: Individual commodity returns, roll yield, seasonal factors

Derivatives: Underlying price, implied volatility (by strike and tenor), rates, dividends

P&L Calculation Methods

Full revaluation: Reprice every position under the stress scenario using the full pricing model. Most accurate but computationally expensive. Required for options and structured products with non-linear payoffs. Sensitivity-based (Greeks): Approximate P&L using first and second-order sensitivities: P&L approximately = Delta dS + 0.5 Gamma dS^2 + Vega d_sigma + Theta dt + Rho dr Fast and suitable for portfolios dominated by linear positions. Accuracy degrades for large moves or highly non-linear positions. Factor model approximation: Express portfolio return as a linear combination of factor returns and apply stressed factor values: R_p = sum(beta_k * R_k) + epsilon Under stress scenario: P&L = sum(beta_k * R_k_stressed) -- ignores the residual term, which is usually appropriate for systematic stress events.

Reporting and Governance

Stress test results should be presented in a standardized report:

Summary table: Scenario name, total P&L, P&L as percentage of NAV, P&L as multiple of daily VaR

Attribution: P&L contribution by asset class, strategy, and top 10 individual positions

Comparison to limits: Whether the stressed loss exceeds any risk limits or capital thresholds

Action items: Recommended portfolio adjustments if stressed losses are unacceptable

Sensitivity: How results change if scenario parameters are varied by plus or minus 20%

Advanced Techniques

Coherent Stress Testing

Ensure that stressed risk factor moves are internally consistent. If the scenario specifies a 30% equity decline, interest rates should decline (flight to quality) unless the scenario explicitly models a stagflation event. The covariance structure of risk factors during historical crises provides a guide for ensuring coherence. Conditional stress scenarios: Specify a primary shock (e.g., equities -20%) and derive the conditional expected moves of other risk factors using the conditional distribution: E[X_2 | X_1 = x_1] = mu_2 + rho (sigma_2 / sigma_1) (x_1 - mu_1) This produces scenarios where all risk factors move in a manner consistent with their historical co-movement during stress.

Probabilistic Stress Testing

Assign probabilities to scenarios to compute expected losses: Expected Stress Loss = sum(P(scenario_k) * Loss(scenario_k)) This bridges the gap between stress testing (scenario-specific) and risk measurement (distribution-based). Scenario probabilities are inherently subjective but can be informed by historical frequencies, market pricing (options-implied probabilities), and expert judgment.

Machine Learning-Enhanced Scenarios

Use generative models (GANs, VAEs) trained on crisis-period data to generate novel stress scenarios that share statistical properties with historical crises but represent new combinations of risk factor moves. This addresses the limitation of historical scenarios (finite set of past crises) while maintaining plausibility.

Key Takeaways

Stress testing complements VaR and ES by assessing portfolio vulnerability to extreme but plausible scenarios that may not be captured in the recent return distribution

Historical scenario replay applies real crisis dynamics to the current portfolio; hypothetical scenarios test vulnerabilities to unprecedented events; reverse stress testing identifies the most dangerous plausible scenarios

Full revaluation provides the most accurate stress test results for non-linear positions, while sensitivity-based approximations are sufficient for portfolios dominated by linear exposures

Coherent stress testing ensures that risk factor moves are internally consistent, using conditional distributions to derive secondary factor movements from primary shocks

Stress tests should be run at least monthly, with results reported to risk committees including P&L attribution, limit comparisons, and recommended adjustments

Frequently Asked Questions

How often should stress tests be updated?

Stress test scenarios should be reviewed quarterly and updated whenever portfolio composition changes materially. The stress test process itself should run at least monthly for institutional portfolios, and weekly or daily during periods of elevated market stress. Regulatory stress tests (CCAR/DFAST for banks) are annual, but internal risk management benefits from more frequent testing.

How many scenarios should I run?

A core set of 8-12 scenarios is standard: 5-6 historical crises (selected based on relevance to the current portfolio's risk profile), 3-4 hypothetical scenarios (targeting known portfolio concentrations), and 1-2 reverse stress tests. Additional ad hoc scenarios should be developed in response to current market conditions or specific portfolio changes.

Can stress tests replace VaR?

No. Stress tests and VaR serve complementary roles. VaR provides a daily risk estimate under normal conditions, suitable for position limits and capital allocation. Stress tests assess vulnerability to extreme events that are outside the scope of VaR models. A portfolio can have an acceptable VaR but unacceptable stress test results, indicating hidden tail risk. Both are necessary components of a comprehensive risk management framework.

How do I stress test a machine learning trading model?

Beyond portfolio-level stress testing, ML models require model-specific stress tests. Test the model with out-of-distribution inputs (market data unlike the training set), adversarial inputs (designed to exploit model weaknesses), and regime-change scenarios (where the statistical relationships the model learned may break down). Monitor model confidence scores during stress scenarios: a well-calibrated model should report high uncertainty for inputs far from its training distribution.

Should stress test results affect position sizing?

Yes. If a stress test reveals that a particular position or concentration would produce losses exceeding the portfolio's risk tolerance under a plausible scenario, position sizing should be adjusted. The standard approach is to calculate the "stressed risk budget" -- the position size that would keep stressed losses within acceptable limits -- and use the minimum of the standard risk budget and the stressed risk budget for actual sizing. Our Position Size Calculator turns that stressed risk budget into an actual share count from your account size, risk percentage, entry and stop.

Stress Testing Portfolios: Historical and Hypothetical