The Next Credit Cycle in Housing..., 23 Jul 2016

Although predicting its timing can be difficult, most would agree that future recessions will occur. So too with credit cycles. We cannot predict exactly when, why, or the severity, but we can be rather confident predicting that another peak in credit risk will occur. In our most recent analysis, the arrows are pointing toward the start of another cycle...

Following the analysis techniques published recently by myself and Jose Canals- Cerda, the Prescient Models team gathered data from Fannie Mae and Freddie Mac for a similar analysis. Combining both data sets to predict default rate, an Age-Period-Cohort model was created to capture the default life cycle versus age and environment function versus time. Then using available origination attributes (FICO, LTV, property type, purpose, channel, occupancy, first time home buyers, term, DTI, etc.), an origination score was created to predict monthly default rate given lifecycle and environment as inputs.

In addition to the base model, dummy variables for vintage were added. This allows us to capture adverse selection, defined loosely here as performance not explainable by observable attributes. In technical terms, we added a fixed effect for vintage to our model.

The following graph shows both models. The solid black line is the predicted credit risk by vintage from observable factors. The blue dashed line incorporates adverse selection. The high-risk vintages of 2006-2008 are explainable both in lower underwriting standards and higher adverse selection. From 2009, dramatic changes were made loan acceptance criteria, causing a sudden reduction in credit risk.

The narrative so far is rather well studied. Our current interest is in why it might be happening now. Notice that starting with Q1 2013, credit risk started to rise, both from observable factors and adverse selection. Although systematic improvements have appeared in FICO and LTV through most of the analyzed time period, the last two years have seen rising LTV, rising DTI, and more first-time home buyers.

Since adverse selection includes risk not in the provided observables, we were particularly interested in what else might be happening with loan origination. The following graph shows just the adverse selection present in each vintage. Here again, we appear to be off the lows of 2011 and 2012 and trending toward higher risk.

Fannie Mae and Freddie Mac are not the entire mortgage industry, but I doubt the rest of the industry is trending to less risk when Fannie and Freddie are trending higher. In fact, the opposite is likely true. We cannot precisely call the peak of the next credit cycle, but the end of 2012 looks like a good marker for the end of the Great Recession credit cycle.

This looks like a good time to be mindful of the underwriting policy choices being made.

New Product Development...

At Prescient Models, we predict the future. It's that simple.

But getting acceptance of those predictions is not always a simple thing. We wanted to create a user interface for our models, but knew that it must be much more than graphs and charts.

Introducing, PrescientManager(tm). This is far more than a data viewer. It's a complete model management system. Of course, we start with a home page that has the forecasts and important portfolio diagnostics, but this package focuses on validation and documentation far beyond anything else on the market. With each user-controlled model update, a full suite of validation tests is updated. All of the model components, forecasts, and validation results are viewable on demand and downloadable as a complete pdf report.

The model update process has been carefully designed with careful consideration of the validation and examination process. Lifecycles, macroeconomic sensitivities, and scoring coefficients are held fixed between formal reviews, but portfolio monitoring of adverse selection and non-economic environmental shifts is continuously updated.

PrescientManager supports the full range of Prescient's models. Simple macroeconomic models, roll rates, vintage models, and loan-level models are all manageable through the same intuitive interface.

Let us know when you're ready for a demo.

Prev Blog Next