Enhanced Transitive Stress Testing: The Dynamic Difference

In reality, a stress test is only as good as its working parts. Tests that rely on an insufficient number of factors, or include inputs that are too focused on specific past disruptions, for example, may leave firms overexposed to an entire realm of unforeseen hazards.

In a previous blog on the different types of stress tests available to asset managers, we considered transitive (or predictive) stress testing, including best practices for creating forward-looking scenarios in order to identify asset-class correlations. But what if we were to add an enhanced or “dynamic” version of the stress test, one that uses mixture models that allow the test to implicitly select the proper blend of evaluations and inputs? By dynamically adjusting a regime’s impact and monitoring the subsequent core factor shifts, we can then determine the likelihood that a given regime was in fact responsible for the resulting shocks.

When used in conjunction with traditional transitive testing, this dynamic approach can provide managers with a powerful and more precise prognosticating tool than “static” or user-defined scenarios.

Finding the Right Regime

Where transitive testing typically uses subjective historical data for volatility and correlation estimation, the dynamic stress test takes this one step further. Through a mixture model (specifically a logistic mixture of linear components model, or LMLC, which uses past probabilities for modeling criteria) we can create a number of different test scenarios covering a range of market environments, from periods of relative tranquility, to intermediate or more pronounced volatility. This allows the test to dynamically adjust to changes in shock size, and we can then use probability weighting to link factor shocks to the “culprit” regimes.

Bear in mind that the “regime-switching” approach within the dynamic stress test may not be suitable in all instances. For example, the “pull-to-realized effect” found within dividend-futures contracts—where volatility decreases as the maturity date approaches—results in a fixed structural shift, rather than a return to an earlier regime. Accordingly, knowing when—and when not—to adjust probabilities based on the dynamics of a particular asset class is key to proper stress testing.

Toward a Better Stress Test

While the markets continue to lurch forward with nary a hiccup, astute managers nonetheless remain poised to take action should fortunes suddenly change. Given the wealth of factors that could trigger a sustained pullback, it is simply not enough to limit one’s risk assessment to a conventional set of historical assumptions.

Transitive stress testing—which uses diverse forward-looking scenarios to identify volatility arising from asset-class correlations—represents a solid approach to achieving better P&L forecasting. Rather than relying solely on subjective insight, however, managers can go one better by including a logistics-based mixture model, which allows the test itself to select applicable periods of volatility, while using an even broader range of shocked scenarios to further refine our outcomes. We believe this dynamic approach offers risk-management teams a level of insight that is essential to building a truly reliable stress-test regimen.

To learn more, read "Toward Dynamic Stress Tests" here.



Scientists. Software engineers. Researchers. At Axioma, we bring together the best minds from across the world to challenge the way financial institutions think about risk and investment portfolio management. We innovate. We create. We collaborate. Together, we deliver leading risk management and portfolio construction solutions to asset managers and hedge funds around the world.