What is the best way to handle missing data in credit scoring models?

1 Why missing data matters

Missing data can occur for various reasons, such as human errors, system failures, non-responses, or incomplete records. Depending on the type, pattern, and amount of missing data, it can introduce bias, noise, or uncertainty in your model, leading to poor performance or misleading results. For example, if you ignore or delete missing values, you may lose important information or reduce the sample size. If you impute or replace missing values, you may introduce artificial correlations or distort the distribution of the data.

3 How to prevent missing data

The best way to handle missing data is to prevent it from happening in the first place. You can do this by improving the quality and consistency of your data collection, processing, and management processes. For example, you can design clear and user-friendly forms, validate and verify the data inputs, use consistent and standardized formats and codes, and monitor and audit the data regularly. You can also communicate and educate your customers and stakeholders about the importance and benefits of providing complete and accurate data.

4 How to impute missing data

If you cannot prevent missing data, you can impute or replace it with reasonable estimates. There are many methods and techniques to impute missing data, depending on the type, pattern, and amount of missing values, as well as the nature and purpose of your model. Some of the common methods are: mean, median, or mode imputation, regression imputation, nearest neighbor imputation, multiple imputation, or machine learning imputation. You should choose the method that best preserves the characteristics and variability of the data.

5 How to evaluate missing data

After you handle missing data, you need to evaluate the impact and quality of your solution. You can do this by comparing and testing the performance and results of your model with and without missing data, or with different methods of handling missing data. You can also use metrics and indicators to measure the accuracy, reliability, and interpretability of your model, as well as the fairness and compliance of your credit decisions. You should aim to minimize the error, bias, and uncertainty caused by missing data.

6 How to document missing data

Finally, you need to document and report how you handle missing data in your credit scoring model. You can do this by creating and maintaining a data dictionary, a data quality report, and a model documentation. These documents should include the information and rationale behind your data sources, definitions, variables, methods, assumptions, results, and limitations. You should also disclose and explain how you handle missing data to your customers and regulators, especially if it affects their credit scores or outcomes.

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?