Each month, we get new Notices of Default (NODs) filed at county courthouses. Our NOD database, always expanding, comprises the outcome against which we can see, at any time, how predictive our scores have been — or how to modify them to further boost effectiveness.

We begin the analytics process by dividing properties by type (use): restaurants, shopping centers/malls, offices, etc., across 95 different use categories.

In each category, we regularly re-compute a recent two-year default rate and its variance against a national average rate.

This process requires careful setup. For instance, historically, restaurants have had higher-than-normal risk. But the main risk is in the independent restaurant category, so it helps to be able to separate chain restaurants like McDonalds from the rest.

Risk by use is a good start, but only a start.

Property value change plays a role, too, just as it did in housing from 2007 to 2011. We look at property value changes utilizing a metro area value change index.

Local geographic concentrations of risk are important as well. We utilize a proprietary database of local area risk concentrations.

We bring in external data, such as certain tax lien data.

Of course, prior commercial mortgage defaults by an existing owner are highly predictive. Less obviously, defaults on the same property under prior owners also can be a problem, potentially indicating something wrong with a building’s location or parking.

On existing mortgages, vintage plays a role, and so do mortgage terms. Building age, condition, and size can play a role.

So does the composition of tenants or owner occupants.

A tenant in business for 30 years will typically be a much more stable payer than a startup firm. That’s pretty obvious, so it’s having the data that counts. We compute the average time-in-business for all tenants at a property where tenant data are available.

You may agree that all the items just listed are clear and logical risk predictors. Let’s flip to the results.