author-image

By

View All Reference Kits

Background

US lenders issue trillions of dollars in new and refinanced mortgages every year, bringing the total mortgage debt to high levels year after year. In 2021, six million homes were sold in the US1, with the total value of new and refinanced mortgages reaching almost $4 trillion2. At the same time, mortgage delinquencies usually represent a significant percentage, and represent a huge debt risk to the bearer. For a financial organization to understand its risk profile, it is critical to build a good understanding of default risk. Organizations are using AI models to help with risk analysis. However, these models come with their own set of complexities. With expanding and changing data, they must be frequently updated to accurately and timely capture the current environment.

In a Fannie Mae* survey from 20183, lenders identified the top three most valuable benefits of AI and machine learning to be:
 

  • Better detection of fraud or defects in the underwriting process
  • Better prediction of borrower default risk
  • Better prediction of borrower prepay or refinance risk

These represent the more lengthy aspects of loan underwriting process. 

According to the McKinsey Global Institute, AI and machine learning could generate over $250 billion4 in value for the banking industry by improving decision-making and risk management, and by tailoring services.

In Q4 2021, mortgage delinquencies were 4.65%5 and outstanding balances of unpaid principals was approximately $2.6 trillion6, which results in approximately $120 billion in mortgage debt at risk in one quarter. Also in 2021, the average time to complete the foreclosure process was 941 days7, costing lenders an average of $50,000 per home8. Considering that there were 151,153 foreclosure filings in 20219, the result is approximately $7.6 billion in foreclosure costs alone.  

This kit offers enterprise financial organizations the opportunity to implement AI to better manage loan default risk, incorporate alternative datasets faster, handle larger datasets, and service their customers faster with reduced underwriting processing wait time. The resulting business value to the financial organization includes lowering the cost of loan delinquencies and processing applications.

This reference kit uses Intel® software products to optimize the training and inference of loan default prediction models. Faster model training improves customer service quality, speeds up loan processing, and enables lenders to simultaneously process more customers.

Solution

In collaboration with Accenture*, Intel developed this loan default risk prediction AI reference kit that may assist your application to increase the efficiency of loan processing and default risk assessment. This reference kit includes:
 

  • Training data
  • An open source, trained model
  • Libraries
  • User guides
  • Intel® AI software products

This reference kit provides an AI model with Intel® Optimization for XGBoost* to predict the probability of a loan default using client characteristics and the type of loan obligation. This reference kit includes:
 

  • A reference end to end (E2E) architecture using an XGBoost classifier
  • An optimized reference E2E architecture enabled with Intel® Optimization for XGBoost* and daal4py


At a Glance

  • Industry: Financial services
  • Task: Train and use an AI model using XGBoost to predict the loan default probability based on client characteristics and the type of loan obligation
  • Dataset: Simulated loan data 
  • Type of Learning: Supervised learning, binary classification
  • Models: XGBClassifier
  • Output: Predicted probability for client to default on a loan
  • Intel® AI Software Products:
    • Intel® AI Analytics Toolkit (AI Kit)
    • Intel Optimization for XGBoost 

Technology

Optimized Intel® AI Software Products for Better Performance.

Intel AI Analytics Toolkit (AI Kit)—Achieve end-to-end performance for AI workloads.

Intel Optimization for XGBoost 

Intel has been directly upstreaming many optimizations to provide improved performance on Intel® CPUs. This well-known, machine learning package for gradient-boosted decision trees now includes seamless, drop-in acceleration for Intel® architectures to significantly speed up model training and improve accuracy for better predictions.

Performance was tested on Microsoft Azure* Standard_D8_V5 using 3rd generation Intel® Xeon® processors for optimized performance. 

Benefits

For the financial organization application developer, this reference kit is optimized for machine learning training and inference while unlocking additional compute capacity. This reference kit offers the enterprise financial organization the opportunity to manage risk, incorporate alternative datasets faster, handle larger datasets, and service their customers faster with reduced underwriting processing wait time. The resulting business value to the financial organization is lower cost of loan delinquencies and processing.

This kit may be useful to optimize credit models within organizations that extend credit as part of their product offerings, such as car makers and farm equipment dealers. There is potential value in using this kit for portfolio credit optimization to help improve risk management models. AI techniques are applied to quantify and minimize credit risk exposure via loss distribution and value-at-risk metrics.

Download Kit

Related Reference Kit

More Efficient Credit Card Fraud Detection

Additional Resources

Intel® AI Software Portfolio

Intel AI Analytics Toolkit

References

  1. US Existing Home Sales 2005-2023, Statista Research Department, June 27, 2022, https://www.statista.com/statistics/226144/us-existing-home-sales/#:~:text=In%202021%2C%20the%20U.S.%20home,and%20increase%20again%20in%20202
  2. 2022 State of the Mortgage Industry Half-Time Report, August 11, 2022, https://blog.mortgagecoach.com/2022-state-of-the-mortgage-industry-half-time-report
  3. Mortgage Lender Sentiment Survey, Fannie Mae, Q3 2018 Topic Analysis, published October 4, 2018.
  4. Derisking Machine Learning and Artificial Intelligence, February 19, 2019, McKinsey & Company, https://www.mckinsey.com/capabilities/risk-and-resilience/our-insights/derisking-machine-learning-and-artificial-intelligence 
  5. Mortgage Delinquencies Decrease in the Fourth Quarter of 2021, Mortgage Bankers Association (MBA), February 10, 2022, https://www.mba.org/news-and-research/newsroom/news/2022/02/10/mortgage-delinquencies-decrease-in-the-fourth-quarter-of-2021-x288743 
  6. OCC Mortgage Metrics Report, Office of the Comptroller of Currency, Fourth Quarter 2021, page 5, published March, 2022, retrieved October 6, 2022, https://www.occ.treas.gov/publications-and-resources/publications/mortgage-metrics-reports/files/pub-mortgage-metrics-q4-2021.pdf
  7. US Foreclosure Activity Drops to an All-Time Low in 2021, ATTOM*, January 13, 2022.
  8. How much does it cost a lender to foreclose on a property? Luke Arthur, Sapling.com, https://www.sapling.com/8625068/much-cost-lender-foreclose-property
  9. Foreclosures Hit Record Low Thanks to Covid Relief. Will It Continue? Natalie Campisi, Forbes Advisor, updated January 20, 2022, https://www.forbes.com/advisor/mortgages/foreclosures-hit-record-low/