In recent times, though, a new form of centralized control has taken root–one that is the work not of old-fashioned autocrats, committees, or rule books but of statistical models and algorithms. These mechanistic decision-making technologies have value under certain circumstances, but when misused or overused they can be every bit as dysfunctional as a Muscovite politburo. Consider what has just happened in the financial sector: A host of lending officers used to make boots-on-the-ground, case-by-case examinations of borrowers’ creditworthiness. Unfortunately, those individuals were replaced by a small number of very similar statistical models created by financial wizards and disseminated by Wall Street firms, rating agencies, and government-sponsored mortgage lenders. This centralization and robotization of credit flourished as banks were freed from many regulatory limits on their activities and regulators embraced top-down, mechanistic capital requirements. The result was an epic financial crisis and the near-collapse of the global economy. Finance suffered from a judgment deficit, and all of us are paying the price.
This is from an article based on a forthcoming book. As I understand it, the thesis is that substituting computer models for human judgment reduces diversity and thus has some of the same costs as substituting central planning for decentralized trial and error.
In my view, it is romantic to think that mortgage lending was decentralized before there was credit scoring. Mortgage lending standards could not be diverse as long as Freddie and Fannie were such dominant players. At Freddie Mac, we were never going to knowingly take the credit risk on a mortgage that was not underwritten to our standards. (We might put a loan that did not comply with our standards into our securities, but only if the loan originator agreed to take the losses if the loan defaulted.) So the question was never one of decentralized judgment vs. centralized standards. It was a question of centralized standards articulated as underwriting guidelines or centralized standards articulated as credit score limits.
Of course, loan originators were always trying to push the envelope on the standards, and that produced some diversity that was not intended on the part of Freddie and Fannie. Trying to get rid of documentation requirements was one envelope-pushing tactic, which Freddie and Fannie shut down in 1991.
Then, under different leadership a dozen years later, Freddie and Fannie opened the door to low-docs. Thus, the advent of NINJA loans–No Income, no Job, no Assets, which did not mean that borrowers were not required to have such attributes. It only meant getting rid of the requirement that borrowers produce copies of tax returns, letters from employers, or bank statements to allow lenders to verify the income, job, and assets being reported on the loan application. In 1991, we thought that eliminating documentation requirements was an invitation to fraud. A dozen years later, the new geniuses at Freddie and Fannie looked at it as a simplification of the lending process. That point of view only cost taxpayers a couple hundred billion dollars and counting.
If the credit scoring approach had a weakness, it was that it led to overconfidence, perhaps leading the mortgage industry to believe that they could get away with NINJA loans. But to the extent that the mortgage market suffered from a monoculture, that reflected the market structure in which two firms dominated. In fact, that dominance was challenged by the growth of subprime lending, which initially took place outside of Freddie and Fannie, who managed to recapture their overwhelming market share at exactly the wrong time in 2006 and 2007.
Bhide’s thesis that computer models create a monoculture is interesting, and I will want to read the book, but I am skeptical about the application to mortgage lending. Keep in mind that I suffered through a lot of corporate soap opera twenty years ago arguing for credit scoring as a better tool than rules of thumb, and so I may not be unbiased.
READER COMMENTS
Tracy W
Aug 17 2010 at 11:51am
This didn’t strike me as an impressive essay. There have been financial crisises before. Was the Dutch Tulip crisis, or the South Sea bubble, or the 1870s crash, or the Great Depression caused by overly-centralised models? Perhaps, but Amar Bhide doesn’t explore this question at all.
Hyena
Aug 17 2010 at 12:13pm
Wasn’t mortgage lending more decentralized in the past, though? One official rationale of Fannie and Freddie was that they would nationalize the mortgage market, allowing investors in the nation’s financial centers to make loans-by-proxy by purchasing securities based on them. That seems to be an indicator that the mortgage market was intensely decentralized.
agnostic
Aug 17 2010 at 3:47pm
Whether or not they lead to more monoculture, they definitely reduce the amount of information that the parties involved are inspecting.
Compression isn’t necessarily bad, as it makes the market more liquid — it’s easier to spread a story that’s “the gist” rather than the whole book — but it also serves the interest of frauds when the environment allows them to flourish (for whatever reasons). They trade in grade-inflated items, and by the time others figure it out, the damage is already done.
So where we suspect hucksters are widespread, we should try to decompress the information. Go back to the boots-on-the-ground evaluation of mortgage borrowers. Stop using GPA for college admissions and look more at the quality of their work that got them that GPA — grade inflation gives B’s to shoddy work on dumbed-down assignments.
No one would decompress unless they have the incentive to, and when fraud pays they clearly do not. So this all boils down to draining the swamp we’ve created that allows frauds to flourish.
Steve Sailer
Aug 17 2010 at 4:27pm
It would seem like we had a national monoculture before — 20% downpayment — and eroded.
FICO was a good thing, much more resistant to fraud than income figures because FICO is measured by disinterested third parties. FICO and downpayments were a good system. FICO, low doc “stated income”, and no downpayments were a very bad system.
david
Aug 18 2010 at 3:31pm
I see ill-considered quantitative modelling essentially as a subset of behaviours flowing from moral hazard. The use of models emerged to “justify” bad loans that would probably not have been made based on local judgment and traditional metrics.
I think in the absence of moral hazard, one would see banks relying far less on models because many of the risks being assessed would not be precisely quantifiable.
Tracy W
Aug 20 2010 at 8:03am
David – how do you explain financial crises before the invention of computers?
Comments are closed.