Click image to show analysis steps...×
This tour show how similar lenders' portfolios are across time, FICO, LTV, and DTI. Using complex graph networks and machine learning, this visualization walks through the evolution of the mortgage markets pre- and post-bubble.
Selecting the year 2000, lets look at the distribution of channels across the graph. The darker lines are the critical path lines based on a minimum spanning tree algorithm, and the lighter gray line are the other top connnections for a certain portfolio.
As we can see channels (BROKER, CORRESPONDENT, RETAIL) are broadly distributed across the graph with no distinct clusters.
In contrast to channels, loan purpose (CASH-OUT REFINANCE, OTHER REFINANCE, PURCHASE) has clear similarity and clustering. The clusters in 2000 have relatively few connections across clusters, which highlights the distinct characteristics of credit risk across all lenders by loan purpose.
The clusters are becoming more connected to each other, but still relatively distinct.
The number of connections between clusters has drastically increased, which illustrates the increased similarity across loan purpose within lenders' portfolios. In fact, the two formerly distinct refinance clusters appear as one.
By 2006, the mortgage market seems to have condensed into one cluster where all types of loan purpose have similar credit risk characteristics.
Notice how one purchase loans portfolios is more align to refinance portfolios than to other purchase loans portfolios, and several other refinance portfolios scattered through the network. Notice how many connections from the former clusters are now across clusters.
In 2007 the clusters are starting to move apart, albeit very slowly.
Loans orginated in 2008 have more diversity in credit risk than the past two years, as the number of connections between clusters continues to diminish.
By 2009, the only connections between clusters are the sole critical paths. This shows that there is a high level of credit diversification between loan prupose clusters, more than likely from drastic changes in underwriting standards across the industry. In addition, notice how much larger the refinance bubbles are than purchase bubbles, highlighting the fact that new purchase loans are being outpaced by refinance loans.
Trying to revitalize the mortgage market, their is an increase in incentives for refinancing, which causes more similarity between the refinance clusters.
As the refinances start to slow down, the similarity between the two finance clusters spreads back out.
By 2013, there are very few connections between the two refinance clusters, and only the critical path connection between the purchase and refinance clusters. Once again, the credit risk of portfolios are highly distinct across different loan purposes.