Recorded: April 28 | 1:00 p.m. ET
Outliers and anomalies refer to various types of occurrences in a time series. Spike of value, shift in level or volatility or a change in seasonal pattern are common examples. Anomaly detection depends on specific context.
In this month's installment in our Data and Machine Learning Workshop Series, RiskSpan Co-Founder & CIO Suhrud Dagli is joined by Martin Kindler, a market risk practitioner who has spent decades dealing with outliers.
Suhrud and Martin explore unsupervised approaches for detecting anomalies.
Co-founder and CIO, RiskSpan
Managing Director, RiskSpan
[pardot-form height=”700 ” id=”6173″ title=”April 28 Workshop: Anomaly Detection”]
Recorded: April 21 | 1:00 p.m. ET
Manually tuning MBS prepayment models is messy. In what amounts to an elaborate trial-and-error exercise, modelers must frequently resort to subjectively selecting sub-populations to calibrate, running back-testing to see where and how the model is off, and then tweaking knobs and re-running the back-test to see the impacts. Rinse and repeat.
RiskSpan’s Janet Jozwik and Steven Sun present an approach for running a set of back-tests on MBS pools that automatically solves for the right set of tuners to align model results to actuals. Learn how, by automatically covering every feasible combination of model knobs possible, you can visualize for every pool the impact each knob combination has on:
- Modeled prepay vs. actuals
- Model error
- Refi incentive and other pool features
Managing Director, RiskSpan
Director, RiskSpan
[pardot-form height=”700 ” id=”6171″ title=”April 21 Webinar Recording: Automated Prepayment Model Calibration Using Machine Learning”]
Recorded: March 31 | 1:00 p.m. ET
Traditional statistical models apply a single set of coefficients by pooling a large dataset or for specific cohorts.
Hierarchical models learn from feature behavior across dimensions or timeframes.
Suhrud Dagli and Jing Liu host an informative workshop applying hierarchical models to a variety of mortgage and structured finance use cases, including:
- Changes in beta and covariance of portfolios across time
- Loan performance across geographies and history – e.g., combining credit performance data from 2008 with unemployment-driven credit issues in 2020.
- Issuer-level prepayment performance
Co-founder and Chief Innovation Officer, RiskSpan
Jing Liu
Model Developer, RiskSpan
[pardot-form height="700 " id="6052" title="March 31 Workshop: Advanced Forecasting Using Hierarchical Models"]
On January 21st, RiskSpan senior managing director Bill Moretti moderated a panel at IMN’s Non-QM Virtual Forum.
The discussion, entitled “Underwriting Credit Standards, Assessing Ability to Pay & Evaluating Default Risk: Are You Protecting Yourself Against a Second Wave or Going All Out?” featured a cross-section of industry participants. These included a rating agency (DBRS Morningstar), a wholesale originator (Oaktree Funding Corp.), an aggregator/securitizer (Annaly Capital Management) and two technology companies (RiskSpan and LoanScorecard), each of whom obviously approaches the underlying question from a slightly different perspective.
Underwriting standards, of course, are best explored in the context of evaluating credit risk. Unprecedented market disruption in response to Covid clearly laid bare the fact that many (if not most) mortgage borrowers are not in a financial position to endure a significant curtailment to income for any sustained period. The discussion focused on what this means for new-origination underwriting standards after Covid.
The panel tackled a number of questions, including:
- How is the market assessing risks in new loans given the current market conditions?
- How are income declines considered?
- What role does forbearance play?
- Should underwriters be looking harder at borrower net worth or taking other assets into account?
- Should PPP payments be counted as borrower income?
- Should DSCR calculations be revisited or modified for investment-property borrowers?
- Is the market taking alternative and other non-traditional data into account when assessing credit risk?
A review of non-QM performance in 2020 relative to that of prime and GSE (CRT) loans revealed some interesting insights:
- Non-QM loans reached a significantly higher 30-day DQ peak (13% in April and May) compared to 4% and 3% for prime and CRT loans, respectively. An analogous gap was observed prior to Covid, however.
- 60+ day DQ (including BK/FC/REO) showed something similar with non-QM loans peaking at 14% in June/July compared with 4% and 5% for prime and CRT loans, respectively. These rates improved by the end of the year but remain elevated relative to pre-Covid levels. The panelists attribute this to forbearance activity.
- CDRs began trending upward in July and August, reaching 10 bps across all classes. CRT CDRs spiked to 50 bps in September and October before retreating to 40 bps, while prime and non-QM CDRs ramped up to the 30 bps and 40 bps, respectively, by year-end.
- CPRs among non-QM loans held steady throughout the year at around 25%. Prime and CRT loans experienced a steady increase in CPR, rising from 20%-25% earlier in the year to approximately 45% by the end of the year. Record refinancing activity appeared to drive much of this.
Partially in response to Covid, rating agencies have implemented changes in their requirements for structural features, including the following:
- Waterfall changed from pro–rata allocation to full sequential pay
- Triggers eliminated based on DQ, loss and credit enhancement
- DQ P&I advancing assumptions reduced from 4-6 months to 0-3 months
- Credit enhancement increased, especially among lower–rated tranches — excess spread and reserve account requirements are now required
- Rating scenario analyses and stress assumptions increased to reflect Covid forbearance assumptions
The pandemic created a significant disruption for non-QM origination and acquisition. Some market participants struggled to adapt during the pandemic’s early months as funding and margin calls impacted mortgage buyers. Aggregators who maintained strong relationships with their originators had more success maintaining funding commitments. These relationships were critical to maintaining overall market health and liquidity. An ability to adapt, allow forbearance and modifications, and work with borrowers was viewed as equally important.
How is underwriting likely to evolve?
The panelists agreed that a revamped underwriting process must consider different sources of income as well as borrower assets and reserves.
Additional borrower requirements that have been proposed include the following:
- Consistency of income in the borrower’s work industry
- Additional weight accorded to other assets held by a borrower
- Consideration of PPP loans as borrower assets but not income
- Stricter underwriting for DSCR loans including a required reserve
- Additional scrutiny of loans made to foreign nationals
The pandemic had an outsized effect on non–QM origination volume, which continues to experience headwinds from Agency-eligible production. There’s just no getting around the fact that, all else equal, brokers find it more profitable (not to mention easier) to focus on Agency production. The importance of specialization continues to be felt, as non–QM aggregators tend to focus more of their attention and efforts on “pure” Non–QM origination shops, as opposed to full-service mortgage bankers, which originate a mix of Agency and non-Agency mortgages. Non-QM underwriting standards will likely need to take this reality into account.
What role will technology play?
While not yet as ubiquitous as in Agency lending, front-end automated underwriting systems continue to make strides in the non-QM world. This growth in AUS consistency and efficiency is a critical component to creating a digital environment for mortgages and accelerating the approval process while maintaining strong risk management and compliance.
The industry is crying out for a clean end-to-end loan acquisition solution for aggregators and other whole–loan portfolio investors. Investors are increasingly looking to get into whole loans, but the secondary whole–loan acquisition process is extremely demanding from an operational perspective. RiskSpan’s Edge Platform enables residential whole–loan buyers to outsource many of these functions.
Beyond whole loans, non-QM securitization data and analytics continues to be a source of angst, due in part to inconsistent forbearance and modification reporting. RiskSpan is seeking to alleviate these pain points by working with clients to standardize and normalize reporting inconsistencies, particularly in the non-QM space. The goal is to provide a way for investors and other market participants to benchmark deals against one another, confident that delinquency, forbearance, and modification comparisons are truly apples-to-apples.
Finally, the panel discussed industry efforts to improve clarity around mapping (or bucketizing) loan types. Doing this is challenging in the non–QM sector because there are so many (literally hundreds) different types of loan documentation requirements. Understanding these is vital to modeling credit risk. Mapping time series data based on the loan type is hard, RiskSpan is at the forefront of developing methodologies to speed and simplify analysis by logically mapping many different loan types into fewer buckets.
Contact us for a free demo and to discuss how RiskSpan can combine its powerful Edge Platform with expert services to help you tackle your thorniest underwriting data and modeling challenges.
RiskSpan is thrilled to be sponsoring IMN’s Non-QM Virtual Forum on Thursday, January 21, 2021. RiskSpan Senior Managing Director Bill Moretti will be moderating a panel at 2:50-3:35 PM.
Click HERE to view the agenda along with details on Bill’s panel: “Underwriting Credit Standards, Assessing Ability To Pay & Evaluating Default Risk: Are You Protecting Yourself Against A Second Wave Of Coronavirus Or Going All Out?”
Bill will be joined at the forum by a team of RiskSpan executives and other leaders, including CEO Bernadette Kogler, CBO Maulik Doshi, and managing directors, Pat Greene, Fowad Sheikh, and David Andrukonis. They will be available for virtual meetings throughout the day.
Recorded: January 13, 2021 | 1:00 p.m. ET
Traders and investors rely on time series patterns generated by asset performance to inform and guide their trading and asset allocation decisions. Economists take advantage of analogous patterns in macroeconomic and market data to forecast recessions and other market events.
But you need to be able to spot these patterns in order to use them.
Catch the latest in RiskSpan’s series of machine learning and data workshops as Chirag Soni and Jing Liu, two of RiskSpan’s experts working at the intersection of data science and capital markets, demonstrate how advanced machine learning techniques such as Dynamic Time Warping and KShape can be applied to automate time series analysis and effectively detect patterns hiding in your data.
Chirag and Jing will discuss specific applications, explain popular algorithms, and walk through code examples.
Join us on Wednesday, January 13th!
[pardot-form height=”700″ id=”4972″ title=”January 13 Workshop: Pattern Recognition in Time Series Data”]
Recorded: Dec. 2nd | 1:00 p.m. EDT
RiskSpan Director Steven Sun shares a procedural approach to tackling the difficulties of efficiently extracting structured data from images, scanned documents, and handwritten documents using Google’s latest Document AI Solution. This approach greatly improves:
- Effectiveness and accuracy of extracting data which will be otherwise difficult or impossible, and
- Automating and streamlining the process of feeding extracted data into a data analytic framework
[pardot-form height=”700″ id=”4909″ title=”Webinar Recording: December 2 Workshop: Measuring and Visualizing Feature Impact & Machine Learning Model Materiality”]
Director, RiskSpan
Watch RiskSpan CEO Bernadette Kogler’s interview with Acting Comptroller of the Currency Brian Brooks.
They discuss many topics include the OCC’s Project REACh, machine learning models to expand the credit box, blockchain’s role in housing finance, and the expanding definition of a chartered institution.
Recorded: Oct. 28th | 1:00 p.m. EDT
RiskSpan CIO Suhrud Dagli, who discussed how ML is being incorporated into model risk management during our Sep. 30 webinar: Machine Learning in Model Validation, demonstrates in greater detail how machine learning can be used:
- In input data validations,
- To measure feature impact, and
- To visualize how multiple features interact with each other
[pardot-form height=”700″ id=”4705″ title=”Webinar Recording: Measuring and Visualizing Feature Impact & Machine Learning Model Materiality”]
Co-Founder & Fintech Lead, RiskSpan
Recorded:
Sep. 17th | 1:00 p.m. EDT
RiskSpan’s Bernadette Kogler led a panel of industry experts in a review of the U.S. economy and how mortgage companies can employ alternative data to responsibly extend mortgage credit more broadly to current and potential homeowners.
Participants include
- Bernadette Kogler, Co-Founder & CEO, RiskSpan
- Amy Crews Cutts, President, AC Cutts and Associates
- Janet Jozwik, Managing Director, RiskSpan
- Laurie Goodman, Director, Housing Finance Policy Center, The Urban Institute
