A randomized experiment in auto lending reveals that algorithmic underwriting outperforms human underwriting, with 10.2% higher profits and 6.8% lower default rates.
While both perform similarly for low-risk loans, human underwriters struggle with riskier, complex loans, unlike machines, which maintain stable performance across all loan types.
As the finance industry increasingly leans into automation, the role of machine learning and algorithms in underwriting is becoming ever more prominent.
A comprehensive study titled “Rise of the Machines: The Impact of Automated Underwriting” by Mark Jansen, Hieu Quang Nguyen, and Amin Shams explores this transformation within the auto lending sector.
The study, which was published on April 30, 2024 in the journal Management Science, offers a detailed comparison between human and automated underwriting, shedding light on the efficiency, profitability, and risk management of each approach.
Study Overview
The study leveraged a randomized experiment conducted by a major auto lender, comparing human and machine underwriting over a large dataset of approximately 140,000 auto loans.
The loans, sourced from over 4,000 U.S. car dealerships, were equally divided between human and machine processing.
This experimental setup provided a robust framework to measure the relative performance of each method without external biases.
Key Findings
One of the most significant findings was the difference in loan profitability.
Loans processed by machines were 10.2% more profitable than those handled by human underwriters.
Notably, the median profitability of machine-underwritten loans matched the 92nd percentile of human-underwritten loans, indicating superior performance by the automated system.
6.8% Lower Default Rates
The study also revealed that machine-underwritten loans had a 6.8% lower default rate compared to human-underwritten loans.
This difference was more pronounced in the early stages of the loan term, with machine-underwritten loans showing 6.9% and 8.3% lower early default rates in the first 24 and 36 months, respectively.
Pricing and Risk Management
Automated systems set higher interest rates, with machine-underwritten loans having an average Annual Percentage Rate (APR) 44.2 basis points higher than those underwritten by humans.
Despite the higher APRs, machine-underwritten loans were less risky.
This indicates that machines priced loans more accurately, taking into account the risk factors more effectively than human underwriters.
Reasons for Superior Machine Performance
Several factors contributed to the superior performance of automated underwriting:
- Consistency Across Loan Types: Machines maintained stable performance across various risk dimensions and loan characteristics, while human performance declined significantly for riskier and more complex loans.
- Agency Conflicts: Human underwriters often approved loans just below critical underwriting thresholds, like the loan-to-value (LTV) ratio of 125%, to maximize their chances of winning the loan auction.
This behavior led to higher default rates for human-underwritten loans in this category.
In contrast, machines, devoid of such conflicts, made more objective decisions.
- Capacity for Analyzing Complex Data: Machines excelled in handling complex data.
The profitability of human-underwritten loans decreased with increasing complexity, whereas machine performance remained stable.
This was particularly evident in loans with low credit scores and high debt-to-income ratios.
Implications for the Auto Lending Industry
The transition to automated underwriting systems in the auto lending industry promises several benefits:
- Increased Efficiency: Automated systems can process loan applications faster than humans, providing quicker decisions that benefit both lenders and borrowers.
- Reduced Bias: Automated underwriting eliminates human biases, ensuring fairer loan approvals based on data rather than subjective judgment.
- Cost Savings: By reducing the need for human underwriters, financial institutions can save on labor costs, further boosting profitability.
The Role of Human Underwriters
Despite the advantages of automated systems, the study acknowledges the importance of human expertise in certain scenarios.
Human underwriters bring a level of discretion and nuanced understanding that can be crucial in borderline cases or unique borrower situations.
A hybrid approach, where machines handle straightforward applications and humans focus on more complex decisions, may offer the best of both worlds.
Performance at Critical Cutoffs
The study found that human underwriters were more likely to approve loans just below the 125% LTV cutoff, often resulting in riskier loans with higher default rates.
Machines, however, priced these loans more conservatively, leading to higher profitability and lower default rates.
For instance, loans just below the 125% LTV cutoff underwritten by humans had a 38.1% higher default rate in the first 36 months compared to those underwritten by machines.
Complexity Handling
Loan complexity was another area where machines outperformed humans.
The study measured complexity based on the expected magnitude of forecast errors in predicting defaults.
Machine-underwritten loans showed relatively stable profitability across different complexity levels, while the profitability of human-underwritten loans declined with increasing complexity.
Conclusion
The research by Jansen, Nguyen, and Shams underscores the transformative potential of automated underwriting in the auto lending industry.
While machines demonstrate superior performance in terms of profitability, default rates, and risk management, the role of human underwriters remains vital in handling complex and unique cases.
The future of auto lending likely lies in a hybrid approach that combines the efficiency and objectivity of machines with the discretion and expertise of human underwriters.
As the industry continues to evolve, financial institutions that effectively integrate automated systems into their underwriting processes will be well-positioned to enhance their profitability and competitiveness in the market.
The rise of the machines in underwriting is not just a trend but a significant shift towards a more efficient and data-driven future in financial services.
