Distressed loans are risky investments. The underlying properties face serious financial challenges and the corresponding loan is often valued at $0.50 on the dollar by the time someone decides to make the investment.
Automating the evaluation process cannot eliminate that risk. But as we discovered by building a credit risk model for one of our large institutional distressed CMBS investor clients, we can use automation to help CMBS investors, brokers and lenders make better, informed decisions and enable far broader deal coverage.
The project began by leveraging Markerr’s location scoring indexes, from sources including payroll data from ADP, real-time permit data, and property level data; data from the client; and Government and Open Data such as the U.S. Census and the Bureau of Labor Statistics (BLS).
Next, our client provided 100 loans across several property types (i.e. Retail, Industrial, Multi-family) to run through the risk assessment model. Additionally, we ran a live parallel process and conducted backtesting.
Our model used machine learning (ML) technology to anticipate NCF (net cash flow) decline by unifying Markerr’s alternative data resources with client data and publicly available data. We built separate models by property type (i.e., Multi-family/Hotel, Retail/Office and Industrial), and applied a number of factors to evaluate the property against an individual loan, such as annual revenue when the loan was underwritten, address and class type; as well as loan characteristics like DSCR (debt service coverage ratio), interest rate, term and the loan-to-value ratio.
The initial results of the model show this methodology can significantly enhance the loan evaluation process. Specifically, two key findings demonstrate the success of the program:
- High accuracy: Our models achieved better than 90% accuracy when compared to historical loan portfolio data provided by our client:
- 93% accurate for Multi-Family/Hotel
- 96% accurate for Retail/Office
- 97% accurate for Industrial
Additionally, 10 loans we flagged as the riskiest of the pool provided significant guidance as to where the client should spend their time.
- Faster results: Our client described the results as highly accurate, and estimated saving their analysts 1-2 weeks of full-time analysis.
Put simply, achieving these levels of accuracy and speed via traditional manual data collection and analysis processes is virtually impossible. With Markerr’s Credit Risk Model, our client is now able to ramp up the number of loans they evaluate while relying on this model as an effective predictor of risk – a powerful combination in the Covid-19 era.
With accurate, quick and efficient assessments of risk, multiple players in the CMBS loan market, including loan investors, lenders, CMBS originators and rating agencies, can mitigate uncertainty during a time of extreme volatility.