ALM Simulation Models: Monte Carlo vs. Scenario-Based

In asset-liability management (ALM), simulation models project how interest rate movements affect the economic value of assets and liabilities over time. Two dominant approaches are Monte Carlo simulation and scenario-based simulation. Monte Carlo simulation uses stochastic-path techniques-also called stochastic simulations-to generate hundreds or thousands of possible future interest rate paths based on probabilistic assumptions (e.g., mean reversion, volatility). Scenario-based simulation, by contrast, constructs a limited set of economically coherent, forward-looking paths-often anchored to historical episodes, stress events, or macroeconomic forecasts-to evaluate discrete strategic alternatives or solvency outcomes.

Monte Carlo simulation in ALM relies on an Economic Scenario Generator (ESG) to simulate a large number of stochastic interest rate paths, typically using dynamic term structure models. Each path represents a possible evolution of short-term rates, yield curve shapes, and volatilities over the projection horizon. The ALM model then revalues assets and liabilities under each path, producing a distribution of outcomes for metrics such as economic value of equity (EVE) or net interest income (NII). This distribution enables the calculation of confidence intervals, Value-at-Risk (VaR), or expected shortfall for interest rate risk.

• Stochastic modeling foundation: Uses calibrated ESGs to simulate paths consistent with observed market data and economic theory (e.g., Cox-Ingersoll-Ross or Hull-White models).
• Outcome distribution: Produces a full probabilistic output, supporting risk metrics that quantify tail exposure and expected losses.
• Use in practice: Commonly applied in internal model validation, capital allocation, and dynamic EAR modeling for budgeting and strategic planning.

Scenario-based simulation constructs a small set-often 3 to 10-of plausible, internally consistent macroeconomic and interest rate paths. These scenarios may be historical (e.g., 1979-1982 tightening), hypothetical (e.g., rapid inflation resurgence), or forecast-based (e.g., consensus macro outlook). Each scenario is applied deterministically to the ALM model, yielding point estimates of EVE or NII under each path. The approach emphasizes interpretability and narrative coherence over statistical precision.

• Narrative-driven design: Scenarios are selected to reflect credible stress or strategic inflection points, often aligned with supervisory expectations or internal risk appetite frameworks.
• Decision support: Used to assess strategic trade-offs-such as asset repositioning or hedging decisions-under specific macroeconomic regimes.
• Integration with governance: Facilitates buy-in from non-technical stakeholders by grounding analysis in clear, story-based assumptions.

Monte Carlo and scenario-based simulation serve complementary roles in ALM, each with distinct strengths and limitations.

• Coverage vs. clarity: Monte Carlo provides broad probabilistic coverage, capturing tail risk and statistical uncertainty, but its outputs can be abstract and difficult to map to specific actions. Scenario-based simulation offers clear, actionable insights but may underrepresent low-probability, high-impact events not embedded in the selected scenarios.
• Model risk: Monte Carlo is sensitive to ESG calibration and distributional assumptions; errors in volatility or mean-reversion parameters can distort risk estimates. Scenario-based simulation is vulnerable to selection bias-omitting plausible paths or over-relying on historical analogs that may not repeat.
• Regulatory alignment: Supervisors (e.g., the OCC) recognize both approaches, but scenario-based methods are often preferred for supervisory stress testing and strategic reporting, while Monte Carlo supports internal model validation and dynamic risk measurement.

Suppose an institution wants to assess the impact of a 200-basis-point parallel upward shift in rates over two years. In Monte Carlo simulation, the ESG generates 5,000 paths where the short rate follows a mean-reverting process with calibrated volatility; the resulting distribution of EVE changes might show a 5th percentile loss of $120 million and a 95th percentile gain of $45 million. In scenario-based simulation, a single, deterministic path-matching the 200-bp shift and its implied yield curve dynamics-is applied, yielding a point estimate of a $90 million EVE decline. The Monte Carlo output supports capital and VaR calculations; the scenario output supports board-level discussion of strategic responses, such as accelerating asset repricing or adjusting duration targets.

• Use Monte Carlo simulation when estimating risk metrics requiring statistical rigor-e.g., economic capital, VaR, or expected shortfall-or when evaluating the impact of uncertainty on long-term strategic outcomes across a broad set of possible paths.
• Use scenario-based simulation when communicating risk to governance bodies, testing strategic decisions under specific macroeconomic regimes, or aligning with supervisory stress testing frameworks where narrative coherence and actionability are prioritized over probabilistic completeness.

Both methods are often used in tandem: Monte Carlo for internal model validation and risk quantification, and scenario-based analysis for strategic decision-making and regulatory reporting.