Expected vs Unexpected: Bridging ECL and Economic Capital in Stress Testing Frameworks

[#1]

Financial stress testing has become a central element of modern risk management, extending well beyond its original role as a regulatory compliance exercise. Today, stress testing informs provisioning decisions, capital planning, risk appetite setting, and senior management decision-making. Within this expanded role, two distinct yet closely related modeling frameworks come into play: Expected Credit Loss (ECL) models and Economic Capital models.

ECL models are designed to quantify the losses a financial institution expects to incur under forward-looking macroeconomic conditions, including adverse scenarios. Economic Capital models, in contrast, focus on the losses that are unexpected and threaten the institution’s solvency. While both frameworks rely on common risk drivers and similar analytical foundations, they are often developed and interpreted independently.

Loss Distributions in Credit Risk

At its core, credit risk modeling revolves around the concept of loss distribution. This distribution reflects the range of potential credit losses that may occur over a given horizon, driven by probabilities of default, loss severities, and exposures. From this distribution emerge two distinct measures of risk – the expected loss, which represents the average outcome, and the unexpected loss, which reflects variability around that average.

ECL models focus explicitly on the mean of the loss distribution. They answer the question of how much loss an institution should reasonably expect, given a particular economic outlook. Economic Capital models, on the other hand, focus on the tail of the distribution, typically at very high confidence levels. They are concerned with low-probability, high-severity outcomes that could jeopardize the institution’s financial stability.

Despite these different objectives, ECL and EC models share a common analytical foundation. Both rely on PD, LGD, and EAD as core inputs, and both are influenced by macroeconomic conditions. The divergence arises from differences in modeling horizons, calibration philosophies, and the interpretation of results. Stress testing plays a critical role by imposing a consistent macroeconomic narrative across these shared inputs, allowing expected and unexpected losses to be analyzed within a unified framework.

Expected Credit Loss Models Under Stress

Under IFRS 9, ECL represents a forward-looking estimate of credit losses that incorporates macroeconomic information and borrower-specific characteristics. In a stress testing context, ECL is typically evaluated under severe but plausible scenarios.

Stress scenarios affect ECL primarily by increasing point-in-time default probabilities, accelerating credit deterioration, and worsening loss severities. As macroeconomic conditions deteriorate, borrowers migrate to weaker credit states, resulting in a higher proportion of exposures.

From a modeling perspective, the mechanics of stressed ECL remain relatively straightforward. Point-in-time PDs are estimated as a function of macroeconomic variables, and these stressed PDs feed directly into ECL calculations alongside stressed LGDs and EADs.

This gives us the loss value under stressed conditions for the entire portfolio, and we can generate N possible portfolio loss outcomes:

This set of values approximates the probability distribution of losses, and we estimate the mean to get:

This calculation captures the increase in expected losses conditional on the stress scenario. However, ECL remains a measure of the mean outcome under stress. It does not explicitly quantify the potential for extreme outcomes. As a result, ECL alone provides an incomplete picture of risk under severe stress.

Economic Capital Models and Tail Risk Under Stress

Economic Capital models address this gap by focusing on unexpected losses. Expected Capital represents the capital required to absorb losses at a high confidence level, often corresponding to a once-in-a lifetime event.

Stress testing has a profound impact on Economic Capital because under adverse economic conditions, defaults become more correlated, recovery rates become more uncertain, and loss distributions develop heavier tails. These effects can cause Economic Capital requirements to increase sharply, even when expected losses rise more modestly.

Most banks define Economic Capital as:

Where,

𝔼[Loss] is the Expected Loss (ECL)

VaR[Loss] is Value-at-Risk at (1-α )×100 % Confidence Level. VaR is the maximum expected loss over a given time period, at a given confidence level. It represents the tail risk of the loss distribution – the “bad but plausible” outcome.

And α is the level of significance.

For example, if we simulate 10,000 scenarios and sort the Losses associated with each of these scenarios in ascending order then we have an ordered loss distribution:

At a confidence level of 99.9%, the VaR is identified as the loss corresponding to the 99.9th percentile of the ordered loss distribution, that is, the 9,990th observation when 10,000 simulated losses are ranked in ascending order. (Note: 99.9 x 10,000 = 99,900)

In other words,

The resulting capital estimate reflects losses deep in the tail of the distribution. Unlike ECL, which responds primarily to changes in average credit quality, Economic Capital is highly sensitive to correlation assumptions and systemic stress.

The correlation referred to here captures the degree to which borrowers’ probabilities of default move together in response to common macroeconomic conditions. A higher correlation implies that defaults are more likely to occur simultaneously across borrowers, leading to a greater concentration of losses. This “fattens” the tail of the loss distribution and results in a significant increase in Economic Capital.

Systemic stress refers to economy-wide adverse conditions that affect many borrowers simultaneously. It refers to shocks that makes defaults more dependent on each other and less diversified. Examples include recessions, housing market crashes and so on.

A Simple Illustration

Let’s look at how ECL, VaR and EC look on a loss distribution using a synthetically generated credit portfolio consisting of 10,000 observations, where each record represents an individual credit exposure.

For each exposure, three macroeconomic variables are simulated: GDP growth, unemployment rate, and interest rate, reflecting prevailing economic conditions. Each exposure is assigned to one of three portfolio segments – Retail, SME, or Corporate – based on predefined probabilities – resulting in an approximately 40% Retail, 30% SME, and 30% Corporate composition.

A latent credit risk score is constructed as a linear function of the macroeconomic variables, where higher GDP growth reduces risk, while higher unemployment and interest rates increase risk. This score is transformed using a logistic function to obtain the PD, ensuring PD values lie between 0 and 1.

Finally, a binary default indicator is simulated by comparing each exposure’s PD with a uniform random draw, producing realized defaults that are probabilistically consistent with the estimated PDs.

Select any image to see a larger version.
Mobile users: To view the images, select the "Full" version at the bottom of the page.

The figure above presents the simulated distribution of total portfolio losses under a stress scenario and illustrates how ECL, VaR, and Economic Capital are derived from this distribution. The histogram represents the frequency of portfolio losses obtained from repeated simulation runs, while the overlaid smooth curve provides a visual approximation of the overall shape of the loss distribution.

The dashed vertical line near the center of the distribution corresponds to the ECL, which is the mean of the simulated loss outcomes. This point represents the average loss that the portfolio is expected to incur under the stressed macroeconomic conditions and is typically interpreted as the level of loss that should be covered through accounting provisions.

The dotted vertical line positioned in the far-right tail of the distribution marks the 99.9th percentile VaR. This percentile indicates an extreme but plausible adverse outcome, such that only 0.1% of the simulated scenarios result in losses exceeding this level.

Economic Capital is implied by the distance between the VaR and the ECL. This difference represents the amount of capital required to absorb unexpected losses beyond the level already covered by provisions. In this way, the figure links ECL and Economic Capital and demonstrates that they are derived from different regions of the same underlying loss distribution.

Linking ECL and Economic Capital Through Curve-Based Stress Testing

The true value of stress testing lies in its ability to connect expected and unexpected losses through a single, coherent scenario. When the same macroeconomic stress is applied consistently across ECL and Economic Capital models, the results provide two complementary perspectives on risk.

Conceptually, stress testing shifts the entire loss distribution. The mean of the loss distribution increases, reflecting higher expected losses, while the distribution also widens and becomes more skewed, reflecting heightened tail risk. Financial analysts often focus on one of these dimensions in isolation, but stress testing demonstrates that both are consequences of the same underlying economic stress.

SAS Stress Testing Solution supports this approach by enabling common data, scenario management, and workflow governance, while allowing distinct analytical methods to coexist. In a typical stress testing cycle, ECL and EC emerge at different stages of this pipeline, but both are outputs of the same stressed portfolio evolution. This shared lineage is the foundation for alignment.

In this section the sample data included in the SAS Stress Testing Solution is used to establish the connection between ECL and Economic Capital. This is accomplished through the curve methodology by translating macroeconomic scenarios into portfolio-level loss behavior.

At the segment level, macroeconomic scenarios are first mapped to PD, LGD, and EAD curves across the projection horizon. The tables below display the ECL, VaR and Economic Capital under baseline conditions. (Note: The tables shown below correspond to the Baseline scenario. In the Adverse scenario section, graphs will be presented to illustrate alternative ways of displaying results from the Economic Capital Analysis node.).

Under baseline conditions (as displayed in the tables above), the fitted curves produce relatively stable or declining PD and LGD paths, leading to a downward-sloping ECL profile over time. The same fitted curves, when propagated through the portfolio loss distribution, yield lower dispersion and therefore lower Economic Capital.

Under the adverse scenario, the curve fitting process shifts and reshapes these PD and LGD curves. The stress does not only increase their levels but also steepens their slopes in the short term, reflecting higher sensitivity to macroeconomic deterioration.

Higher stressed PD and LGD curves increase the area under the curve, which raises ECL. Furthermore, greater curvature and slope increase cross-borrower dependence and loss volatility, which widens the loss distribution and raises Economic Capital.

Products with more convex loss curves – such as Credit Cards and HELOCs – exhibit larger proportional increases in both ECL and Economic Capital under stress, because their fitted curves respond more strongly to adverse macro conditions. By contrast, Residential Mortgages display flatter curves, leading to more moderate ECL and comparatively lower Economic Capital increases.

The time dimension further reinforces this linkage. The ECL projections typically show the largest adverse uplift in the early projection years, consistent with front-loaded stress in the macro scenario. The Economic Capital projections exhibit a similar temporal pattern but with amplified magnitude, as tail losses react more strongly to the same early steepening of the PD and LGD curves. As portfolios amortize and exposures decline, both ECL and Economic Capital contract.

From a modeling perspective, curve fitting enforces consistency between expected and unexpected loss measures. Instead of calibrating Economic Capital separately, both ECL and Economic Capital are derived from the same stressed loss surface defined by scenario-dependent PD and LGD curves. This prevents inconsistencies in risk signals that can arise when expected losses are estimated using one set of assumptions while tail losses are driven by another, ensuring that moderate ECL outcomes are not paired with high tail-risk capital requirements, and vice versa.

Additional Information

For more information on SAS Stress Testing visit the software information page here. For more information on curated learnings paths on SAS Solutions and SAS Viya, visit the SAS Training page. You can also browse the catalog of SAS courses here.

Find more articles from SAS Global Enablement and Learning here.