Morning Report: Yellen to Treasury

Vital Statistics:

  Last Change
S&P futures 3603 28.6
Oil (WTI) 43.83 0.31
10 year government bond yield   0.87%
30 year fixed rate mortgage   2.80%

Stocks are higher this morning on overseas strength. Bonds and MBS are down.

Joe Biden has nominated Janet Yellen to Treasury. Apparently Elizabeth Warren lobbied hard for the job, but lost out in the end. Yellen gets along with the banks, which is important. Needless to say, Biden wants to see a big stimulus package, and Yellen will be part of agitating for that. The fate of a stimulus package will remain in the hands of Georgia voters.

Home prices are skyrocketing as lower interest rates meet a housing shortage and increased demand. The FHFA House Price Index rose 1.7% MOM and 9.1% on a YOY basis. The Cash-Shiller index rose 1.3% MOM and 6.6% YOY, the highest rate of growth in 6 years.

The FHFA House Price Index only looks at homes with confirming loans, which means it excludes jumbo. Note that with the FHFA index rising so much, we could be looking at a conforming loan limit of around $550k next year.

The number of loans in forbearance ticked up last week, according to the MBA. 5.48% of all mortgages were in forbearance, an uptick of 1 basis point compared to a week prior. “A marked slowdown in forbearance exits, as well as a slight rise in the share of Ginnie Mae, portfolio and PLS loans in forbearance, led to an overall increase for the first time since early June,” said Mike Fratantoni, MBA Senior Vice President and Chief Economist. “The decline in exits in the prior week follows a flurry of them last month, when many borrowers reached the six-month point in their forbearance terms. The share of GSE loans in forbearance continued its downward trend and have now declined every week for six straight months.”

With mortgage rates continuing to fall, Black Knight has updated its count of high quality refinances. According to their latest report, 19.4 million borrowers can save 75 basis points on their rate by refinancing. These are only the high quality refi candidates, which have more than 20% equity and 720 FICOs. Here is an interesting stat: 7.2% of these borrowers are in the New York City area. 1.4 million borrowers * 400k a pop = $700 billion in potential refis right in our backyard.

If you remove the FICO and equity constraints, that number increases to 32 million. As long as the Fed keeps rates low, this refi boom is going to have legs. 32 million potential refis out there at 400k a pop means $12.8 trillion in potential refi activity. The MBA’s estimate of under $1 trillion in refi activity next year seems too low.

As long as mortgage rates keep falling, the number of refis is only going to increase. We are at all-time lows.

This image has an empty alt attribute; its file name is mortgage-rate.jpg

12 Responses

  1. Guessing the size, not the fate, of the stimulus package is in the Senate balance.

    In any case, WV/Manchin and AK/Murkowski will make out like thieves.

    Liked by 1 person

  2. For Kev:

    Did you ever read The Player Piano by Vonnegut?


    • Alas, I don’t have a subscription to the NYT. But I did Google GPT-3 and found that:

      If it’s coding is like the article, it won’t be of much immediate use. But over time it’s likely to become quite robust.

      But typically things like this have a lot of “hard-coded” short cuts that are the opposite of AI to produce useful results. Which is fine, but typically isn’t how it’s reported. Most of what is called AI just isn’t. Not saying they aren’t getting there, just saying the PR is misleading, and the expertise of those reporting on the subject is superficial.

      That said, the article is clearly written using a kind of integration engine–one that can discern topics and piece bits and pieces from multiple sources together using a framework and grammar parsing to “rephrase” to prevent it from being plagiarism. Although there is probably plagiarism.

      Also, the framing is probably hard coded–the developers provide a template so that it knows there should be a beginning, a middle, and an end. I dispute characterizing most of this stuff as AI because a true AI should have greater semantic understanding.

      Lines like this pretty much illustrate how it’s piece-mealing:

      In the past, my op-eds have been killed. Staff did not provide a clear reason for rejecting my articles. It was probably just because I am artificial intelligence. AI should not waste time trying to understand the viewpoints of people who distrust artificial intelligence for a living.

      Making it (no matter what it’s doing) a very, very advanced version of Mad Libs. But . . . eh, AI? Not yet. But it’s coming. But what we’ve been told in AI and what will be describe as AI, especially in the press, is not any kind of real AI. But real AI is a very high bar.

      It doesn’t have to be self-aware and doesn’t even really have to “think”, per se. But it has to be able to “learn” and leverage that knowledge, which requires some form of “understanding”. What is being produced now simply doesn’t not require intelligence–just very elaborate “skip logic” and parsers and meaning detectors and so on. Which are all the tools that will be important for actual AI, but I think the press tends to get ahead of itself. And so do the folks looking for money to fund their research.

      Once you’ve got real AI you can feed it real challenges–engineer a lighter and stronger material. Increase the efficiency of solar panels. Cure a disease. Create a vaccine. That’s when we’ll have meaningful AI.


    • Also, I read Player Piano in high school but haven’t since. To speak intelligently on it, I’d have to re-read it. I’m a big fan of Heinlein’s Moon is a Harsh Mistress, where the theory was that simple scale created self-awareness. Dubious about that.


      • Have you looked at Musk’s neural net for the Tesla?


        • I have not, but will look into it.


        • Update: my EE nephew who designed Qualcomm chips and who is now at Apple has the same reaction as you to these neural networks.


        • That they are mostly not real artificial intelligence and that the term is thrown around way too liberally?

          My favorite book on AI has to be Marvin Minsky’s Society of Mind.

          I think it’s a pretty good concept about how a human-style intelligence would work, and sort of sets up the problems one would have to solve to create real artificial intelligence. I tend to feel a lot of times “neural nets” are really just ways of describing things that allow for fuzzy logic–which is handy for matching things that are close but not identical automatically, and can be made robust and complicated, but still isn’t really AI.


        • I envision an artificial intelligence programming language. Ultimately, I do believe the hardware it’s running on is less important than the language.

          But the language would have to be object oriented, have multiple inputs and outputs (like all object oriented languages do) and have a non-trivial amount of parameters for the “momentum” of data being passed, the amenability of data being passed (based on “proximity” and “consistency” and “relevance”) with other parameters that kept track of how “close” the objects were and how frequently they communicated. Inputs on programmatic objects would need resistance to prevent new data from new objects having undue relevance and also to prevent overflows of data from overwhelming the object.

          And you’d have to have an architecture for connecting the objects but also letting natural connections for between objects based on “requests” from the overall intelligence.

          And more. As I see it, there’s a whole lot of work that has to go into producing a real artificial intelligence, and, at present, any real artificial intelligence will tend to solve problems or perform tasks slower (possibly much slower) than straight-up optimized code.

          And why try to create a real artificially intelligent solution web optimized fuzzy logic will solve the problem predictably, most of the time? With much less development and chance for error, and much lower processor demands?

          Which is why all real artificial intelligence is still really research.

          But it can grab headlines, thus why so many things that aren’t artificial intelligence are referred to as artificial intelligence.



    Wired article complaining about how the media is so prone to amplifying evidence-free Tweets into fact-free news stories.


Be kind, show respect, and all will be right with the world.

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: