When it comes to forecasting MSR cash flows, the practice of creating “rep lines,” or cohorts, of loans with similar characteristics for analytical purposes has its roots in the Agency MBS market. One of the most attractive and efficient features of Agencies is the TBA market. This market allows originators and issuers to sell large pools of mortgages that have not even been originated yet. This is possible because all parties understand what these future loans will look like. All these loans will all have enough in common as to be effectively interchangeable with one another.  

Institutions that perform the servicing on such loans may reasonably feel they can extend the TBA logic to their own analytics. Instead of analyzing a hundred similar loans individually, why not just lump them into one giant meta-loan? Sum the balances, weight-average the rates, terms, and other features, and you’re good to go. 

Why the industry still resorts to loan cohorting when forecasting MSR cash flows

The simplest explanation for cohort-level analytics lies in its simplicity. Rep lines amount to giant simplifying assumptions. They generate fewer technological constraints than a loan-by-loan approach does. Condensing an entire loan portfolio down to a manageable number of rows requires less computational capacity. This takes on added importance when dealing with on-premise software and servers. It also facilitates the process of assigning performance and cost assumptions. 

What is more, as OAS modeling has evolved to dominate the loans and MSR landscape, the stratification approach necessary to run Monte Carlo and other simulations lends itself to cohorting. Lumping loans into like groups also greatly simplifies the process of computing hedging requirements. 

Advantages of loan-level over cohorting when forecasting MSR cash flows

Treating loan and MSR portfolios like TBA pools, however, has become increasingly problematic as these portfolios have grown more heterogeneous. Every individual loan has a story. Even loans that resemble each other in terms of rate, credit score, LTV, DTI, and documentation level have unique characteristics. Some of these characteristics – climate risk, for example – are not easy to bucket. Lumping similar loans into cohorts also runs the risk of underestimating tail risk. Extraordinarily high servicing/claims costs on just one or two outlier loans on a bid tape can be enough to adversely affect the yield of an entire deal. 

Conversely, looking at each loan individually facilitates the analysis of portfolios with expanded credit boxes. Non-banks, which do not usually have the benefit of “knowing” their servicing customers through depository or other transactional relationships, are particularly reliant on loan-level data to understand individual borrower risks, particularly credit risks. Knowing the rate, LTV, and credit score of a bundled group of loans may be sufficient for estimating prepayment risk. But only a more granular, loan-level analysis can produce the credit analytics necessary to forecast reliably and granularly what a servicing portfolio is really going to cost in terms of collections, loss mitigation, and claims expenses.  

Loan-level analysis also eliminates the reliance on stratification limitations. It facilitates portfolio composition analysis. Slicing and dicing techniques are much more simply applied to loans individually than to cohorts. Looking at individual loans also reduces the risk of overrides and lost visibility into convexity pockets. 

RiskSpan’s cloud-native Edge Platform projects prepayment, default, severity and MSR cash flows (income and costs) at the loan level

Loan-Level MSR Analytics

REQUEST A DEMO

Potential challenges and other considerations 

So why hasn’t everyone jumped onto the loan-level bandwagon when forecasting MSR cash flows? In short, it’s harder. Resistance to any new process can be expected when existing aggregation regimes appear to be working fine. Loan-level data management requires more diligence in automated processes. It also requires the data related to each individual loan to be subjected to QC and monitoring. Daily hedging and scenario runs tend to focus more on speed than on accuracy at the macro level. Some may question whether the benefits of such a granular, case-by-case analysis that identifying the most significant loan-level pickups requires actually justifies the cost of such a regime. 

Rethink. Why now? 

Notwithstanding these challenges, there has never been a better time for loan and MSR investors to abandon cohorting and fully embrace loan-level analytics when forecasting MSR cash flows. The emergence of cloud-native technology and enhanced database and warehouse infrastructure along with the ability to outsource the hosting and computational requirements out to third parties creates practically limitless scalability. 

The barriers between MSR experts and IT professionals have never been lower. This, combined with the emergence of a big data culture in an increasing number of organizations, has brought the granular daily analysis promised by loan-level analytics tantalizingly within reach.