How COVID-19 Has Changed Data Analytics and Modeling: Three Ways to Respond to Current Challenges
Major models have revolutionized how the mortgage industry does everything from assess risk to assist borrowers who are having difficulty making their payments. They are used to project interest rates, mortgage rates, house prices, unemployment rates, defaults and prepayments, and other key outputs that determine our business success and viability.
But when it comes to predicting the future, you’d have needed a crystal ball at the start of 2020 to foresee the challenges we’d face by spring. Practically overnight, the COVID-19 pandemic flung us into the middle of a "Black Swan” and models across the mortgage industry were challenged to incorporate the dynamic and unprecedented events. Despite recent exponential growth in data availability, this pandemic reminds us that models are only as good as the data upon which they’re based.
In the last few months, it’s been challenging to make sense of the how the pandemic — followed by Federal Reserve monetary stimulus actions, fiscal policy responses and the granting of forbearances—has affected models and businesses.
Here are three ways to consider and respond to current challenges:
- Analyze Past Data
According to The New York Times, the labor market might have hit bottom after recording a 14.7% unemployment rate in April, but it is too soon to know for certain. Indeed, the June 18 unemployment report does not inspire confidence.
Despite the lack of certainty with variables such as the unemployment rate, we can analyze recent events that threw the United States economy into a tailspin, even if no single past event closely matches what we’re facing today. For example, during the nadir of the Great Recession, the unemployment rate was approximately 10% and there was a 30% decrease in house prices. When unemployment rates spiked in the past, they were usually accompanied by a steep drop in house prices, but that isn’t happening now, at least not yet.
We can also leverage insights from previous catastrophic events. In the aftermath of natural disasters such as the 2017 hurricanes in Florida and Texas, people may have stopped work temporarily but most of their jobs didn’t go away and, except for Hurricane Katrina, most of the downturns were short-lived and followed by a quick recovery.
Past data is far from perfect, but if you rely on the closest data that imitates what you’re currently facing, you will gain valuable perspective.
- Obtain New Data and Use New Analytics to Inform Decisions
Since the first few weeks of pandemic pandemonium, we have gained insights, gathered more data and adjusted our models.
As soon as this pandemic began, researchers offered intelligence around the early performance of loans and forbearance rates as well as how rapidly jobless claims were ramping up. Reports now show that forbearance programs put in place by the CARES Act appear to be helping and, for many borrowers, forbearance is manageable. Since the Great Recession wasn’t that long ago, we’ve also seen how lenders are exhibiting muscle memory when it comes to making decisions and leveraging resources during a crisis.
Some of the ways we’ve used new data and analytics to help inform our decisions include:
- Data from lenders, via surveys and platform providers;
- Real-time figures on consumer behavior (e.g., restaurant reservations, hotel occupancy rates, traffic patterns);
- Loss estimates benchmarked to those produced by others; and
- Information gleaned from models such as Home Value Explorer®, which reliably estimates the value of individual homes.
While delayed reporting due to forbearances on payments for credit cards, cars and student loans adds uncertainty to front-end scoring models, we have leveraged our industry relationships to share forecasting information. It benefits us all to have accurate, updated information on factors such as credit trends, house prices and mortgage losses.
The name of the game is quickly determining model overlays so the outputs from the models are “close” to what you are likely to experience. This has helped us feel comfortable with the model adjustments put in place. We’ve also built in flexibility to adjust our models as the situation changes or new data becomes available, which is critical as the situation remains fluid.
- Learn for the Future
We’ve discovered through this crisis that several models —the ones with machine learning embedded in them — are nimble and better able to adjust when circumstances change rapidly. Machine learning models contain more features and are rebuilt and retuned more often than traditional statistical models that might be rebuilt only once a year.
Despite these advances, some challenges moving forward are clear, particularly those relating to obtaining timely data so we will know which loans are likely to receive forbearance – and for those that do, how they will proceed through the delinquency pipeline.
Uncertainty remains heightened. One canary in the coal mine is the percentage of unemployed workers who will have jobs to which they can return. We have no model to predict this or to estimate the recovery timeline for heavily-impacted small businesses.
While current hardships are enormous, new opportunities exist. Perhaps the greatest is how digital mortgages are being advanced in a major way. One estimate is that this crisis is advancing the digital move by five years – speed is motivated by need.
How do you obtain borrower information without exchanging papers in person? How do you close a loan if the county courthouse is shut down? How do you provide borrowers with their loan documents if you don’t attend face-to-face closings? Because of the pandemic, all the events that once were high touch are now becoming low touch as the industry fast tracks digital solutions.
One of the advantages of our increasingly digitized world is access to more immediate performance indicators and measures. From a big data and analytics standpoint, all this digitalization positions our models for further success. When you digitize data, it becomes machine readable and as you gain more machine-readable information, you can use this data to improve models. Companies with sophisticated data analytics and predictive programs are well-positioned to take advantage of the extra data from digitization.
Bottom line: Better predictions can help the mortgage industry make better business decisions. The major benefit of models is that they can scale in ways humans cannot. But models alone can’t be counted on for predictions in unusual or unprecedented cases. With events like the COVID-19 pandemic, human judgment becomes critically important. You never want to lose sight of the fact that models are there to help human decision-making, not substitute for human reasoning. You always need an adult in the room.
* * *
For lenders and housing professionals looking to better understand the mortgage process amid the COVID-19 pandemic, #HelpStartsHere