• 287 days Will The ECB Continue To Hike Rates?
  • 287 days Forbes: Aramco Remains Largest Company In The Middle East
  • 289 days Caltech Scientists Succesfully Beam Back Solar Power From Space
  • 689 days Could Crypto Overtake Traditional Investment?
  • 694 days Americans Still Quitting Jobs At Record Pace
  • 696 days FinTech Startups Tapping VC Money for ‘Immigrant Banking’
  • 699 days Is The Dollar Too Strong?
  • 699 days Big Tech Disappoints Investors on Earnings Calls
  • 700 days Fear And Celebration On Twitter as Musk Takes The Reins
  • 702 days China Is Quietly Trying To Distance Itself From Russia
  • 702 days Tech and Internet Giants’ Earnings In Focus After Netflix’s Stinker
  • 706 days Crypto Investors Won Big In 2021
  • 706 days The ‘Metaverse’ Economy Could be Worth $13 Trillion By 2030
  • 707 days Food Prices Are Skyrocketing As Putin’s War Persists
  • 709 days Pentagon Resignations Illustrate Our ‘Commercial’ Defense Dilemma
  • 710 days US Banks Shrug off Nearly $15 Billion In Russian Write-Offs
  • 713 days Cannabis Stocks in Holding Pattern Despite Positive Momentum
  • 714 days Is Musk A Bastion Of Free Speech Or Will His Absolutist Stance Backfire?
  • 714 days Two ETFs That Could Hedge Against Extreme Market Volatility
  • 716 days Are NFTs About To Take Over Gaming?
Trade In Counterfeit Goods Hits Half A Trillion Dollars

Trade In Counterfeit Goods Hits Half A Trillion Dollars

The counterfeit market has breached…

Another Retail Giant Bites The Dust

Another Retail Giant Bites The Dust

Forever 21 filed for Chapter…

How The Ultra-Wealthy Are Using Art To Dodge Taxes

How The Ultra-Wealthy Are Using Art To Dodge Taxes

More freeports open around the…

  1. Home
  2. Markets
  3. Other

The Re-Release of the Open Source Mortgage Default Model

I have received some feedback on the open source mortgage default model and data, and consequently took it offline due to the possibility that there may have been a material error (see The Truth About the Banks Has Been Released: the open source spreadhseet edition for the background). I have had it reviewed and reviewed it myself several times in light of the external inputs and am now putting a revised version back online for all to download, modify and or distribute. To give a quick background on what was going on, I received a notice of potential errors from a reader, along with suggestions on how he felt the model could be improved. I checked the SCAP Loss Assumptions file that he sent and although there is a difference of approach followed by us and the reader regarding estimation of loan loss rates, we have found both the methods acceptable. The key difference is that the methodology (as explained below) followed by the contributor requires certain (additional) assumptions to be made while we wanted to refrain from making additional assumptions to remove any subjective biases and base our analysis using Fed's data to the maximum extent possible. For those that do not wish to read through this lengthy history, let it be known that the end result of the reader's input, and to a lesser extent our slightly tweaked methodology both resulted in a HIGHER loss rates for mortgages (and consequently, the banks), not a lower one. So, if the original open source model was to be considered inaccurate, it would be leaning towards the optimistic side, which further drives home my point and thesis concerning the government's bogus stress tests and the current and future state of many US banks (see Welcome to the Big Bank Bamboozle! for the detailed story). Banks and asset managers have rallied enormously again, causing me to recheck my thesis and research, and yet still, I find absolutely no indication of my viewpoint being in error. This situation is similar to the dot.com era where many fundamental types tried to short the internet companies, and many got blown out of the water before the industry and the entire market collapsed. Past is not prologue, but I have been in this minimum vaue view of what I consider overvalued assets and markets when I sold off my real estate in 2006. Patience is hard to come by when you are facing against the wind, but I am not comfortable going against both the math and common sense.

One potential source of the perception of error was the inclusion of Alt-A ARMs resetting after 24 months (the 24+ month column in the Alt-A tab) in the cumulative loan loss rate section for 2 years. We feel that the exclusion of Alt-A ARM resetting after 24 months would underestimate loan losses during the actual 24 month period due to a variety of reasons, the least of which are the facts that once a certain LTV level is breached you will have early loan recasts (see The banking backdrop for 2009 where I warned of this at the beginning of the year) and the fact that there will be some defaults before an initial reset is achieved.

Those who just want to dowload the model click here (Revised SCAP Assumptions Public Open Source Version 1.1 2009-05-18 15:15:47 1.21 Mb), otherwise read on to how I got to the second slight revision of the model.

Although reset date is an important aspect to predict default rates for Alt-A ARMs there are additional factors including current LTV, borrowers paying capacity, % change in property prices, unemployment rates etc that are important determinants to predict default rates for Alt A loans. Hence altogether excluding Alt-A ARM resetting after 24 months would underestimate default assumptions.

I have have highlighted the key difference in the methodology between the approaches below.

Our methodology:

We had computed total Alt A ARM loans to be reset (including loans to be reset after 24 months) and calculated default rates on Alt A loans to be reset based on default rate assumptions considering FICO scores and LTV. We had applied an overall weighted average default rate of 80% for all loans on high risk Alt A loans to be reset.

We have made 3 changes to the document to further refine default assumptions but attempt to avoid the injection of subjective bias into the raw government data, where possible.

  • We have now considered total Alt A ARM loans (including loans already reset) to consider default on Alt A ARM loans.

  • Also to better reflect default rate for each category separately we have now computed separate default rate for each loan category. We have now computed default rates on Alt A loans based on separate default rate assumptions (in the table below) for each loan category defined as - High risk loans (low FICO and low LTV), low risk loans (high FICO and high LTV) and other loans (high LTV / Low FICO score and low LTV and high FICO score).

  • The change has not materially impacted overall default rate for Alt A ARM loans to be reset.

We have further bifurcated the loans into owner occupied and non-owner occupied, since the investor and absentee owner loans have a very high propensity to default when under water, particular in comparison to loans where the owner actually lives in the property. Below are our assumptions for default rates for each category (linked dynamically in the respective sheets). Be aware that I have left the model open, so any and all can feel free to input their own assumptions.

Based on the above assumptions, the weighted average default rate for Alt-A Arm loans is now at 25.33% (21.4%% previously) while weighted average default rate for Sup prime Arm loans is now at 39.19% (41.4% previously).

The overall loss rate has been computed as default date * (1-recovery rate). Recovery rate has been computed taking into consideration LTV at origination, % change in hosing price and historical recovery rates.

High Risk Alt-A ARM Loans
(Low FICO and high LTV)
Low Risk Alt-A ARM Loans
(High FICO and Low LTV)
Avg Risk Alt-A ARM Loans
(High FICO and High LTV / Low
FICO and Low LTV)
Owner Occupied Non-Owner Occupied Owner Occupied Non-Owner Occupied Owner Occupied Non-Owner Occupied
50% 90% 15% 30% 25% 50%
 
High Risk Subprime ARM Loans
(Low FICO and high LTV)
Low Risk Subprime ARM Loans
(High FICO and Low LTV)
Avg Risk
(High FICO and High LTV / Low
FICO and Low LTV)
Owner Occupied Non-Owner Occupied Owner Occupied Non-Owner Occupied Owner Occupied Non-Owner Occupied
65% 95% 20% 35% 35% 60%

The contributing reader's methodology:

The contributor had bifurcated total loans (including loans to be reset after 24 months and loans already reset) into owner occupied ALT-A, non-owner occupied ALT-A, owner occupied ALT-A ARM and non-owner occupied ALT-A ARM. These loans are classified into high risk loans based on reset schedule as per the table below, eg All loans resetting within next 12 months are classified as risky loans.

Pls note : We had classified loans into high risk loans based on LTV and FICO scores and not based on reset schedule which would require subjective assumptions (or alternatively, a more granular and very recent historical data set which we currently do not have access to.

High risk loans in relation to Resetting
Already reset 20%
within 12 mo 100%
in 12-24 mo 50%
24+ mo 15%

After classifying loans into owner occupied/ non- owner occupied and risky / non-risky loans these default rates are computed based on following assumptions.

Default rates
Owner Occupied ALT-A 15%
Non-Owner Occupied ALT-A 30%
Owner Occupied ARM 20%
Non-Owner Occupied ARM 40%
Owner Occupied ARM High Risk 50%
Non-Owner Occupied ARM High Risk 70%

Pls note : We had initially assumed single default rate for owner occupied and non-owner occupied

We appreciate that the reader's methodology could potentially improve the estimation of default rates for Alt-A ARM loan losses by bifurcating Alt A ARM loans into sub category, but we also consider the potential of injecting subjective bias an unnecessary weakness given the breadth of the NY Fed's and the FDIC's data set. We have, however added inputs to differentiate between investor loans and owner occupied loans as stated above.

Overall the different methodology no not provide significant variation in Alt A loans with our Total Alt A loss rate of 20.4%(computed based on Fed's data for delinquent loans, Foreclosure and REO) versus 19.1 % loss rate for Alt A non-ARM by the contributing reader (computed based on assumptions above).

Our default rates for Alt A ARM loans is at 25.33% (up from 21.4% in the original model) versus 32.6% by the contributing reader. The difference is not due to different methodology but due to slightly higher default rate used by the contributing reader (Alt A non-ARM).

 

Back to homepage

Leave a comment

Leave a comment