• Explores how geopolitical risk creates structural breaks that traditional, history-based risk models cannot capture.
• Examines how firms should stress test for fast regime shifts, persistent shocks, and unpredictable policy responses – not gradual downturns.
• Highlights how effective analysis must link geopolitical events to cascading impacts across market, credit, and operational risks.
• Discusses how models should incorporate expert judgment, Bayesian updating, and focus on impact sensitivity over precise probabilities.
• Explores how resilience requires mapping and stress-testing interdependencies between models, including feedback loops that can amplify shocks.

Ahead of Advanced Model Risk USA, we spoke with Krishan Sharma to explore modelling emerging risks and geopolitical risks

Q1. Geopolitical risk has shifted from a tail risk to a core modelling challenge. What’s the biggest limitation you see in traditional risk models when it comes to fast-moving geopolitical shocks?

The Core Limitation: The Reliance on Stationarity and Historical Lookbacks.

The fundamental issue is that our traditional models, whether VaR, Expected Shortfall, or PD/LGD models, are empirically calibrated. They rely on the assumption of stationarity: that the future statistical properties of a system will resemble the past.

Geopolitical shocks are, by definition, structural breaks. A sudden sanction regime or a blockade doesn't behave like a standard deviation move in a normal distribution; it changes the distribution entirely. Traditional models are 'rearview mirror' driven. When a geopolitical shock hits, the correlation matrix often goes to one, and historical calibration windows (even stressed ones) fail to capture the velocity of the transmission. We are trying to model 'unknown unknowns' using a dataset of 'known knowns,' which leads to a dangerous lag in signal detection."

Q2. Scenarios like stagflation, trade fragmentation, and policy volatility are increasingly plausible. How should firms be stress-testing their macroeconomic models to better reflect today’s geopolitical environment?

The Shift: From Severity to "Velocity" and "Persistence."

"We need to move beyond standard regulatory exercises (like CCAR or DFAST) which often assume a linear degradation of variables over 9 quarters. To capture today's environment, firms should:

• Introduce 'Regime-Switching' Scenarios: Instead of just shocking GDP or Unemployment, we must model the relationship changes. For example, in a stagflation scenario, the typical negative correlation between equities and bonds might flip to positive, nullifying diversification benefits.
• Reverse Stress Testing: Instead of asking 'What happens if rates rise 500bps?', ask 'What geopolitical event would break our liquidity position?' and work backward.
• Model 'Policy Whipsaw': Scenarios usually assume a policy response that smooths the shock. We need to stress test for policy errors or volatility—where central bank reaction functions become unpredictable or contradictory to fiscal policy."

Q3. Tariffs, sanctions, and supply-chain disruptions rarely operate in isolation. What approaches are most effective for quantifying their combined impact across market, credit, and operational models?

The Approach: Causal Linkage and Transmission Mechanisms.

"Siloed modeling is the enemy here. A tariff is not just a market risk factor; it is a transmission mechanism. The most effective approach is Integrated Scenario Analysis that maps the causal chain:

1. Input: Tariff announced (Political Event).
2. Market Risk: Immediate spike in commodity volatility and FX dislocation.
3. Credit Risk: This feeds into sector-specific borrower PDs (e.g., manufacturing margins get squeezed).
4. Operational Risk: Simultaneously, this triggers supply chain failure (idiosyncratic op risk).

We are seeing success using Network Theory or Bayesian Belief Networks to map these dependencies. You cannot simply sum the VaR, Credit VaR, and OpRisk capital; you must quantify the cross-risk correlations which tend to spike during these specific events."

Q4. Geopolitical risks are often qualitative and hard to parameterise. How can institutions embed geopolitical sensitivity into models without creating false precision or overconfidence?

The Strategy: Bayesian Overlays and "Champion-Challenger" Frameworks.

"The danger of 'false precision' is real—assigning a 12.4% probability to a trade war is meaningless. Instead, we should:

• Use Qualitative Overlays (IMA): Accept that the core model cannot capture the risk. Apply a formal, governance-approved 'Management Adjustment' or overlay based on expert judgment.
• Bayesian Updating: Use models that allow for prior beliefs (expert sentiment) to be updated with incoming data (news flow), rather than waiting for quarterly GDP prints.
• Sensitivity > Probability: Stop obsessing over the exact probability of the event. Focus on the sensitivity of the portfolio to the event. If a Taiwan strait crisis implies a 40% drawdown, it doesn't matter if the probability is 2% or 5%—the exposure needs hedging. We need to model the impact conditional on the event, not just the likelihood."

Q5. Interdependencies across models can amplify shocks. What practical steps can firms take to improve model connectivity and strengthen enterprise-wide resilience?

The Solution: A Dynamic "Model Inventory Map."

"Most firms have a model inventory, but few have a Model Dependency Map. To improve resilience:

• Map the 'Hand-offs': visually map how Model A (e.g., Interest Rate Macro Model) feeds Model B (Prepayment Model) which feeds Model C (Liquidity Stress Test).
• Stress the Links: Often, the models work fine individually, but the hand-off breaks during stress (e.g., the macro model outputs a negative rate that the credit model's code can't handle).
• Feedback Loops: Implement 'second-order' testing. If Credit Spreads widen (Model A output), does that feed back into the Macro Model (Model B input) to further depress GDP? Most current frameworks are linear; building in these feedback loops is critical for true resilience."