Recently, we have seen a number of articles concerned about an ever-flatter yield curve, which could invert after another couple of Fed hikes, and the historical recession-forecasting abilities of inverted curves. On the other hand, payroll and wage growth, along with core PCE, have remained firm. Forecasts for the amount of net issuance have been revised upward and the Treasury increased auction sizes across the curve in its August 1st quarterly refunding announcement and will be introducing a new two month bill. Meanwhile, the curve has steepened somewhat, with 2s10s spreads now 32bps and the 10-year flirting with the 3.00% level. The statement from the August Fed meeting provided no surprises, leaving the curve plenty of room to re-steepen later in the year after another 1-2 Fed fund hikes, updates on inflation and the continuation of QT and net supply evolution.

In this issue of the Pipeline, we present two articles. Our first article, by Dan Szakallas, discusses the impact of the new tunings for agency LDM that are effective as of July 31, 2018. As Dan shows, agency 15-year collateral, along with 5/1 Hybrids and 7/1 Hybrids, have been showing significant deviation from model projections as rates have been rising. The tuning changes bring performance into line.

Our second article, by Alex Levin, Mickey Storms and John Gao, looks at how our latest release of LoanKinetics (LK Version 1.10) uses a probability-weighted three-scenario approach with discounted cash flows to provide a CECL solution for mortgage holdings. The article provides an overview of LK functionality, addresses period-over-period attribution and the analytical framework, and has examples. The appendices provide a useful summary of LK components as well as graphical CECL-related intermediate calculations by scenario along with results.

At the end of July, we released an updated set of tuning files for Agency LDM versions 2.1.2 and 2.2 within the monthly data file update. These files had been available for testing by clients during the month of July before becoming official as of July 31st. This article discusses the reasoning behind the updates, and also looks at the impact of the changes on both historical data as well as model forecasts.

Specifically, the models affected by the update are the FNMA and FHLMC 15YR fixed-rate models, and the FNMA and FHLMC 5/1 and 7/1 hybrid ARM models. In these models, the generated forecasts were coming in below the actual prepayments observed over the past year. The new tuning values have addressed those issues, bringing model results back in line with observed prepayment trends.

LoanKinetics is AD&Co’s comprehensive, multi-functional, whole-loan application. The most recent release—LoanKinetics 1.10 for CECL—adds new functionality that solves CECL (Current Expected Credit Loss) with a focus on analytical rigor, compliance, and completeness.

LoanKinetics estimates CECL losses using a variation of the discounted cash flow method (DCF) as the expected present value of projected cash flow losses over three probability-weighted economic scenarios. The losses are modeled using our LoanDynamics prepayment and default model (LDM). The system permits CECL loss estimates to be reconciled period-to-period using a built-in loan loss attribution feature.

For reporting purposes, CECL loan losses are aggregated and presented by vintage and credit rating using LPR (Loan Profile Ranking), our proprietary loan credit quality rating. An example of the segmented, expected loss results is demonstrated in this article.

Because LoanKinetics also features COAS (credit- and option-adjusted spread) loan pricing technology that generates loan fair value, the system can solve for CECL and carrying values for the full spectrum of loans held on the balance sheet, including: (1) loans marked at fair value, (2) loans available for sale, (3) loans held to maturity, as well as (4) loans purchased with significant purchase credit deterioration (PCD) since origination or acquisition.