Economic Forecasting Under CECL – A Guide

 

alt_economic_forecasting_under_cecl_a_guide

Introduction

Economic forecasting is one of the most important components for CECL. It requires a more comprehensive understanding of the economic environment than in the past, and the ability to assess potential changes in the future. When determining your forecast, it is important to consider all available data, along with the current economic climate, and to make assumptions that are reasonable and supportable. This guide will provide an overview of the skills and strategies needed to correctly forecast and update your CECL allowance.

Know Your Data and Statistical Significance

It is important to have a strong understanding of all the data that will be used in economic forecasting, as well as the relationships between different data sets. This will ensure that your forecasts are reasonable and supportable. To support this assessment, ARCSys recommends following the process and our statistical analyses, which we complete annually for our clients:

  1. Segmentation Analysis: Accurately assess the risk associated with each segment and ensure that the economic forecasting is based on reliable data. To analyze an institution’s data, one must first ensure that they appropriately segment their assets and that there is enough activity in each segment of assets. Segments may need to be further broken out into subclasses. This also ensures that the collateral codes are accurate and up-to-date. Re-evaluating the risk structure annually will allow for more reliable allowance calculations. 
  2. External Factor Analysis: Once the assets are pooled into their segments, identify the economic indicators that are likely to be statistically significant in relation to the dependent variables and determine which are most relevant. The factors that are statistically significant may change over time and must be re-evaluated annually to support reliable results.
  3. Risk Analysis for Statistical Significance: This takes the economic indicators identified previously and determines, at the pool level, which ones are statistically significant in relation to each dependent variable. It offers insights into the economic indicators that strongly affect charge-off, recovery, prepayment, and credit utilization activities across the observed performance periods.

The CECL model requires a significant amount of data to produce accurate and reliable estimates of credit losses. Historical loss information generally provides a basis for an institution’s assessment of expected credit losses. It is important that financial institutions take the time to carefully select the right data to use in their CECL modeling efforts.

Determine Your Forecast

Develop an expectation for how the market will behave in the future to accurately forecast potential losses. Understanding how factors generally move with other factors is important to the overall forecast. Institutions must use reliable data sources that provide accurate, timely, and sufficient data to make effective forecasts. ARCSys recommends reviewing various forecasts, because each forecast has a built-in bias. Below is a list of sources that are publicly available:

  • FRED
  • Fannie Mae
  • Freddie Mac
  • TD
  • Deloitte
  • S&P Global
  • Federal Reserve
  • Wells Fargo
  • Goldman Sachs
  • The Conference Board
  • Green Street

The Relationship Between Forecasted Variables and Dependent Variables

To best estimate the effect of an external variable on a forecasted dependent variable, determine how that external variable has changed over time. An important component when forecasting for CECL is understanding the relationship between economic indicators and credit risk. For example, generally as the Unemployment Rate increases, the likelihood of charge-offs also increases. If you are forecasting the Unemployment Rate to go from 3.6% to 5%, determine whether the unadjusted historical data includes Unemployment Rates within that range. If the forecasted range is not encompassed within the selected dataset, it may be necessary to make adjustments to the historical dataset or select a different historical range of periods.

Expectations of the Effects of Forecasts on Data

Relying on a result without first setting an expectation of what you think should occur may cause you to accept a result that is incorrect. When beginning to forecast, have your own general expectations about economic conditions. Consider what you hear on the news and in conversation with others, and look at trends in the markets. Forecasted expectations are input based on your beliefs of future economic or employment conditions. Consider to what extent the selected dataset already incorporates forecasted expectations.

The CECL standard (326-20-30-9) requires that an institution should forecast for the periods in which management believes that they can create and document a reasonable and supportable forecast. The length of the forecast term can have a significant impact on the accuracy, sensitivity, and complexity of the forecast and the resulting estimates of expected credit losses. Back-testing can help an institution determine if their selected data and forecasts are producing results that approximate actual activity.

Calculating the Effects

After identifying the external variables that are statistically significant for the selected historical dataset and determining your forecasts, an institution then needs to utilize those independent variables with their dataset. For example, an institution determined that they were going to use the National Unemployment Rate as an independent variable. The dependent variable (charge-offs) and independent variable (Unemployment Rate) are compared with each other through time. To start, this means the change in the dependent variable (net charge-offs) is compared to the change in the independent variable (Unemployment Rate), month-to-month.

Statistical models learn from the comparison of the historical datasets (independent and dependent) and apply the found relationships to create a forecast of the dependent variable, based on where the institution believes the independent variable (Unemployment Rate) will move in the future. Essentially, the institution is making a prediction of how the dependent variable (net charge-offs) will react in forecast periods, based on the learned associations of the past.

It is important to remember that a statistical model can only learn from the data used as an input. For example, if an institution only has data from 2015 to 2023, the model can only learn from those dependent variables and the relationships with the independent variables.

Evaluating the Effects

When viewing the forecast results, one should expect that the forecast will reflect the expectations previously mentioned. When determining if your forecasted dependent variable meets your expectations, consider the following:

  • The forecasted dependent variable should align with your expected losses and credit risk expectations
  • The forecasted dependent variable should reflect management’s knowledgeable view of the economic environment
  • The forecasted dependent variable should reflect any changes in the business environment that could impact its ability to collect on the loans
  • The assumptions used to generate the forecast should be reviewed to ensure that they are reasonable and consistent with your institution’s historical performance and expectations for the future

When reviewing forecast results, it’s important to pay attention to any red flags that may pop up. A red flag would be if your forecasts are significantly higher or lower than your actual historical life cycle loss ranges. That could indicate that something is off with your forecasting methodology or that the forecast provider used to support your forecast has a hidden bias.

If the forecasts do not seem reflective, one of the most important questions to ask yourself is, “Were your expectations for the results reasonable?” There are a few possible reasons your forecasts may not be reflective, if your expectations were reasonable. It could be either the incorrect data was used in the model or that the assumptions made about how the variables would affect the result were incorrect. Understanding the directional effect of the independent variable’s relationship (hazard) on your data is important. If appropriate, consider modifications to the model to better estimate the result. This may mean adjusting input data, assumptions, and model parameters. If the forecast of the external variable goes significantly above or below the actual experience in the dataset selected, an institution may need to change or adjust the dataset.

Documentation

The new CECL standard has raised the bar for what kind of documentation is required, compared to the Incurred Loss Model (ILM) model. Forecasting results need to be displayed in a clear and concise manner. Details should be provided on the following:

  • Data sources used
  • Assumptions made during the forecasting process
  • All forecasted variables and results
  • How the forecast was developed including methodology and any adjustments made

Keeping track of the accuracy of the independent variable forecast over time will give you a clear understanding of how well the model is forecasting.

How ARCSys Supports Throughout the Forecast Process

  • Provides annual statistical updates
  • Annually reviews models with clients
  • Gives the client control over their forecast through detailed inputs of the independent variables
  • ARCSys’ forecasting charts allow clients to see results of the data
  • Provides a Quarterly Economic Update webinar to get most current information

Contact ARCSys to streamline your forecasting process today!

alt_author

About The Author

alt_michael_umscheid

Michael Umscheid

President & CEO