Exploratory modeling uncovers overlooked financial stress scenarios

Exploratory Modelling and Analysis (EMA) introduces a novel framework for macroeconomic scenario analysis under deep uncertainty.

It systematically samples a vast array of uncertain conditioning variables, without imposing strong ex-ante priors on their choice, combination, or distribution.

This approach flips conventional scenario methods by forecasting outcomes from these sampled uncertainties and then tracing policy-relevant results back to the specific variable combinations that generated them.

Applied with an Interacted Vector Autoregression model (I-VARX) to euro area financial stress scenarios, the study reveals several key insights.

Policy-relevant scenario selection is highly model-dependent, with larger or better-fitting models often underrepresenting tail risks, a phenomenon termed the 'fallacy of statistical fit'.

Non-linearities are also crucial for shaping outcomes, emphasizing the need for non-linear models.

The dimensionality of the uncertainty space significantly impacts coverage, as constraining variables risks omitting relevant combinations.

Central banks use narrative or statistical approaches for scenario analysis, each with limitations.

Narrative methods provide structured stories but rely on judgmental scenario selection and point estimates.

Statistical methods offer probabilistic risk assessments but often use simplified models that miss complex system dynamics.

Hybrid approaches, while combining elements, can be constrained by external inputs or may inadvertently dampen extreme tail outcomes.

This paper's Exploratory Modelling and Analysis (EMA) framework addresses these shortcomings.

By extensively sampling conditioning variables under fundamental uncertainty and avoiding strong prior restrictions, EMA systematically identifies policy-relevant scenarios, offering a more comprehensive and flexible method for exploring future states of the world than conventional techniques.