Loss Forecasting

End-to-end loss forecasting for banks and credit unions — from developmental data infrastructure through model development, validation, ongoing monitoring, systematic qualitative adjustment frameworks, and platform development or migration.

Where’s the Value Add?

It’s through Actively Managing Strategic Risks Identified in CECL & Stress Testing

This page focuses on our services and experience, which are extensive, of course. See Credit Regimes, Risk Management, and the Value of Stress Testingg for the non-reporting, non-regulatory benefits of loss forecasting.

Our Specialty

CECL and Stress Testing

Loss forecasting for banks and credit unions — whether CECL or stress testing — is our specialty. That’s end-to-end:

  1. Building and preparing developmental databases (that can later be used for other purposes).
  2. Actual model development and documentation process.
  3. Implementation within commercially-available (or custom) production forecasting platforms.
  4. Ongoing performance monitoring (OPM) programs.
  5. Qualitative adjustment frameworks (systematic, defendable overlay processes) that are closely related to OPM.

We’ve performed this work on portfolios ranging from a few billion dollars to $100 billion.

Can a single loss forecasting approach be best for both CECL and stress testing? Had you asked us in 2018, we would have (and did) say, “no,” but today we know it’s not just possible, it is our standard approach for most portfolios and most regions.1 Our loss forecasting model development approach is based on two innovations:

  1. Loss forecasting based on economic theory, rather than simply identifying statistical regularities, provides rigorous yet intuitive risk measures that are sound, sophisticated, and simple.
  2. Insights into the nature of historical observations and patterns and their applicability to forecasting, i.e., credit losses follow regimes.2

What this looks like in practice: We recently built a complete CECL and stress testing model suite for every major consumer and wholesale portfolio at a $75 billion bank — using a single modeling framework that serves both purposes. We have also built bespoke models for credit unions with non-standard run-off portfolios (solar panel loans, home improvement loans, refinanced student loans) where commercial platforms provide no usable starting point.

Standard practice on every development engagement: We build an independent parallel process in a separate statistical environment to verify all model outputs before they reach a board, an examiner, or a regulator. This is not an add-on service — it is how we work.

We have supported platform migrations from Sageworks to Moody’s ECL and from legacy systems to SAS ECL — including full model rebuilds, not just documentation transfers. We also maintain a proprietary parallel loss-forecasting framework to independently verify vendor platform outputs on every engagement where a black-box platform is in use.

We use AI tools in our development process — generative AI for accelerating data preparation and code development, and machine learning where it genuinely improves model performance. On every engagement, we document where AI was used and where it was not, so the validation team — whether ours or yours — can evaluate each component on its own terms.

Our Approach

Comprehensive Risk Management

Spero Risk Associates has substantial experience with all aspects of stress testing, scenario analysis, and design, especially CECL, CCAR and DFAST. In fact, we built many hypothetical and historical scenarios for both credit and market risk, as well as joint credit-and-market loss scenarios years before the crisis and CCAR. This experience —building enterprise stress testing programs— is the foundation of how we think about loss forecasting: not as a compliance exercise but as a genuine forward-looking risk measurement.

We understand relationships among the firm, its environment, and risk factors, such as the industry, social and political conditions, technology, and regulations, which are as ever-changing as the firm.

Understanding the consequences of potential events can provide valuable, actionable information.

Along with a loss-forecasting framework, we are able to recommend and build appropriate scenario and sensitivity analyses as well as ongoing monitoring to estimate the consequences of potential combinations of events and how forecasts can affect your model’s bounds.

Our Experience

Comprehensive expertise across all major credit risk domains with proven track record in model development and implementation.

Consumer Credit Excellence

Decades of experience building robust models for consumer lending with industry-leading performance metrics.

Real Estate

Mortgages, HELOC, and home equity with property value integration — including residential mortgage model suites built for both CECL and stress testing at a $75 billion bank.

Student and Solar Loans

We built CECL loss forecasting models for solar panel loan, home improvement loan, and refinanced student loan portfolios at a credit union exceeding $25 billion — run-off books with short sample histories where standard methodologies were not applicable.

Credit Cards

Advanced behavioral modeling and utilization rate forecasting.

Automotive Finance

Complete lifecycle modeling from origination to charge-off prediction.

Wholesale Credit Leadership

Specialized expertise in complex commercial relationships with sophisticated risk assessment frameworks.

C&I Lending

Commercial & Industrial loan models with sector-specific adjustments — including C&I model suites built for both CECL and stress testing at a $75 billion bank.

Commercial Real Estate

Multi-family, office, retail, and industrial property loss modeling — including CRE model suites built for both CECL and stress testing at a $75 billion bank.

Bond Portfolios

Corporate bond default prediction and portfolio-level loss estimation.

Non-credit Forecasting

Cutting-edge forecasting methodologies combining economic theory with statistical rigor for superior predictions.

Macroeconomic Variables

Multi-horizon forecasting of regional and national economic indicators — including hypothetical and historical scenario construction at major U.S. financial institutions, built before those frameworks existed in their current form.

PPNR Modeling

Pre-provision net revenue forecasting with fee and expense components.

Operational Risk

Frequency and severity modeling for operational loss forecasting.

  1. The perverse effects of hurricanes (PDs up, LGDs down, and little overall effect) on mortgages means that for stress testing purposes, you might want to use the same functional forms—regression equations—but the underlying sample data used to generate sensitivities—the regression coefficients—might (or should) be different for different purposes. The effects that generate few losses in hurricanes don’t cancel out in forecasts and can substantially, and surprisingly distort results. See our analysis of hurricanes for the reasoning. ↩
  2. See this article for the basic notion, although we discuss regimes in many places on this site. ↩