Mr. Winston Churchill had once mentioned-, 'Never let a good crisis go to waste." A similar notion can be formed about the circumstances of COVID-19. Today, we are dealing with uncertainty at a level that has never been done before. How is forecasting a problem during COVID-19? What can be done to avoid obstacles and move forward in the long run?

In this article, we write about forecasting in a crisis. We take the insights from the presentation of Brett Engel, Director of Finance and Credit risk at Baxter Credit Union, delivered at the 5th Annual CULytics Virtual Summit.

The COVID forecasting problem

Models don’t perform well in tail events (Events which have a small probability of occurring) – A model relies on historical data for information. This pandemic is a tail event, and factors such as unemployment rates, or surplus dollars are not considered while designing forecasting models before such events.

What is needed by Credit Unions is a defendable and actionable forecast so that it is possible to manage it.

Key Steps for Building and Defending an Actionable Forecast

1. Expert Intuition- Seek the intuition of experts to augment model certainty and use it to inform your model.
2. Robust framework- Focus on building a solid forecasting framework and not just the result.
3. Multiple Scenarios- Use macroeconomic forecasts to test a range of potential scenarios. It can be resourceful to rely on external forecasts like Moody's Analytics. Your model should be nimble enough to refresh and update, quickly and easily.

Dealing with Loan Losses during Covid-19- An example of Baxter Credit Union

1. Seeking Expert Intuition- Baxter Credit Union observed unemployment forecasts to be overly pessimistic. They questioned whether it was possible to assess other reputable viewpoints and factor them into the model. They acted quickly to help members with loan extension and to answer how would performance look as a result? These intuitions, formed by organization leaders help guide towards particular directions.

2. Build Framework- The robust and defendable framework starts with building a base model. For example- BCU formed a simple linear regression of its performance at the portfolio level against industry performance to build out forecast losses.

   Adjustments are then made to the base model. For example- Extension Disposition Analysis, Median Unemployment Forecast Adjustment, etc, which eventually lead to the final forecast.

   Everything should be updated monthly.

3. Run Scenarios- Considering the uncertainties being faced, this is the most crucial step. For example-BCU runs its direct auto charge off forecast through multiple scenarios leveraged by Moodys Analytics. These results flow directly into the rolling forecast.

CONCLUSION

The need is to build a defendable and actionable forecast. Actions should be based on this forecast. Leverage expert knowledge to improve forecast models. Intuitive judgments can be modeled by building creative frameworks and finding relevant data sources. The future is unknowable. A range of scenarios should be used to understand the potential impact to your business.