Full disclosure: I’ve known Amar Bhide for roughly 25 years (we both worked on the Citibank account at McKinsey, albeit never on the same project) and although we correspond only occasionally, I continue to regard his as a particularly keen observer and original thinker. He was briefly a proprietary trader, then an associate professor at Harvard Business School (first in finance, later in enterpreneurship), then a professor at Columbia’s business school, and after taking a sabbatical to write a book (A Call for Judgment, due out in two weeks) is joining the Fletcher School of Law and Diplomacy in a newly-established chair.
Amar has an article at the Harvard Business Review which encapsulates some of the core arguments from his book. I’m providing a few extracts here because it appeals to my sensibilities and I therefore think NC readers will like it as well. The most interesting bit to me is the aspect that I highlight in the headline to this post: that the evolution of finance, particularly in its near universal adoption of standardized models in lending processes bears a troubling resemblance in its process and outcomes to a centrally planned economy (funny that people like to dump on the Fed for interest rate setting, and miss the other. widespread aspects of de facto centralization, via standardization and over-reliance on models). Some of his arguments overlap ones I’m made repeatedly here. For instance, I’ve decried the fact that shifting lending from loan officers in branches to standardized, score-based templates resulted in considerable loss of information: face to face assessment of the borrower (does he understand what he is getting into? Does he regard the loan as a serious commitment?) and knowledge of the community (How healthy is his employer? What is the outlook for the local economy?)
Bhide comes to similar conclusions to ones reached here and in ECONNED, and his framing may help finance skepticism get greater traction. From his HBR article:
Because natural laws and mathematical inferences cannot predict behavior, algorithms are built upon statistical models. But for all their econometric sophistication, statistical models are ultimately a simplified form of history, a terse numerical narrative of what happened in the past. (The simplifying assumptions of most statistical models are in fact so great that they can almost never be used successfully to reconstruct the very historical data used to construct the models.) They reveal broad tendencies and recurring patterns, but in a dynamic society shot through with willful and imaginative people making conscious choices, they cannot make reliable predictions….
This doesn’t mean statistical controls and data-mining programs are useless in human affairs. They can debunk false assumptions and stereotypes or suggest new rules of thumb. Faced with a large number of choices (as when thousands apply for one job), they can provide a quick, objective first-cut screen. But predictions of human activity based on statistical patterns are dangerous when used as a substitute for careful case-by-case judgment. They nonetheless continue to gain ascendency. Nowhere has this been more apparent—or more dangerous—than in the financial industry…..
The traditional lending model was built around case-by-case judgment. Home buyers would apply for loans from their local bank, with which they often had an existing relationship. A banker would review each application and make a judgment, taking into account what the banker knew about the applicant, the applicant’s employer, the property, and conditions in the local market. The banker would certainly consider history—what had happened to housing prices, and the track record of the borrower and other similarly situated individuals. But good practice also required forward-looking judgments—assessments of the degree to which the future would be like the past. Dialogue and relationships were also important: Bankers would talk to borrowers to ascertain their beliefs and intentions. And staying in touch after the loan was made facilitated judgments about adjusting terms when necessary….
Over the past several decades, centralized, mechanistic finance elbowed aside the traditional model. Loan officers made way for mortgage brokers. At the height of the housing boom, in 2004, some 53,000 mortgage brokerage companies, with an estimated 418,700 employees, originated 68% of all residential loans in the United States. In other words, fewer than a third of all loans were originated by an actual lender. The brokers’ role in the credit process is mainly to help applicants fill out forms. In fact, hardly anyone now makes case-by-case mortgage credit judgments. Mortgages are granted or denied (and new mortgage products like option ARMs are designed) using complex models that are conjured up by a small number of faraway rocket scientists and take little heed of the specific facts on the ground….
The buyers of securitized mortgages don’t make case-by-case credit decisions, either. For instance, buyers of Fannie Mae or Freddie Mac paper weren’t, and still aren’t, making judgments about the risk that homeowners would default on the underlying mortgages. Rather, they were buying government debt—and earning a higher return than they would from Treasury bonds. Even when securities weren’t guaranteed, buyers ignored the creditworthiness of individual mortgages. They relied instead on the models of the wizards who developed the underwriting standards, the dozen or so banks (the likes of Lehman, Goldman, and Citicorp) that securitized the mortgages, and the three rating agencies that vouched for the soundness of the securities.
Dispensing with judgment has also helped funnel the mass production of derivatives into a few mega-institutions, posing systemic risks that their top executives and regulators cannot control.
Little good has come of this robotization of finance. Reduced case-by-case scrutiny has led to the misallocation of resources in the real economy. In the recent housing bubble, lenders who, without much due diligence, extended mortgages to reckless borrowers helped make prices unaffordable for more prudent home buyers.
The replacement of ongoing relationships with securitized, arm’s-length contracting has fundamentally impaired the adaptability of financing terms. No contract can anticipate all contingencies. But securitized financing makes ongoing adaptations infeasible; because of the great difficulty of renegotiating terms, borrowers and lenders must adhere to the deal that was struck at the outset. Securitized mortgages are more likely than mortgages retained by banks to be foreclosed if borrowers fall behind on their payments, as recent research shows.
When decision making is centralized in the hands of a small number of bankers, financial institutions, or quantitative models, their mistakes imperil the well-being of individuals and businesses throughout the economy. Decentralized finance isn’t immune to systemic risk; individual financiers may follow the crowd in lowering down payments for home loans, for instance. But this behavior involves a social pathology. With centralized authority, the process requires no widespread mania—just a few errant lending models or a couple of CEOs who have a limited grasp of the risks taken by subordinates.
Yves here. This is a particularly succinct indictment of modern finance. It’s unlikely to get the traction it deserves because no new paradigm is waiting in the wings. As Thomas Kuhn argued in his Theory of Scientific Revolutions, scientific (and by implication, intellectual) frameworks persist even as evidence against them mounts, with ever-more patches and work arounds, until a new generation embraces a different paradigm.
But modern finance is a sort of lingua franca, computationally convenient, and most important, a lot of people have businesses deeply embedded in the current way of doing things. And the regulators are just as deeply invested. I found this section of a recent New York Fed paper on the shadow banking system simply astonishing (I’ve wanted to shred other significant elements of this article, but that is a serious undertaking that has to wait a bit). It listed the advantages of securitazation….without offering a list of disadvantages. To wit:
There are at least four different ways in which the securitization-based, shadow credit intermediation process can not only lower the cost and improve the availability of credit, but also reduce volatility of the financial system as a whole.
First, securitization involving real credit risk transfer is an important way for an issuer to limit concentrations to certain borrowers, loan types and geographies on its balance sheet.
Second, term asset-backed securitization (ABS) markets are valuable not only as a means for a lender to diversify its sources of funding, but also to raise long-term, maturity-matched funding to better manage its asset-liability mismatch than it could by funding term loans with short-term deposits.
Third, securitization permits lenders to realize economies of scale from their loan origination platforms, branches, call centers and servicing operations that are not possible when required to retain loans on balance sheet.
Fourth, securitization is a potentially promising way to involve the market in the supervision of banks, by providing third-party discipline and market pricing of assets that would be opaque if left on the banks’ balance sheets.
Yves here. Notice the Panglossian subtext: everything is for the best in this best of all possible worlds of securitization. Not only is there no consideration of the downside, such as the near-impossibilty of dealing with troubled borrowers on a case-by-case basis, but there is no acknowledgement that the same benefits could have been achieved by other, sometimes cheaper, means.
For instance, the first advantage, greater diversification, can be achieved by a less costly route, by selling loans. The second finesses the real problem with mortgages, that the US is pretty much alone among advanced economies in offering thirty year fixed rate mortgages on a large scale basis. Floating rates are the norm elsewhere.
A fixed rate mortgage made sense in a low interest rate volatility environment, but the product continues to exist when its effect is to shift interest rate risk on to banks, who in turn blow up on it periodically (first the savings and loan crisis) and leave taxpayers with losses. So ultimately, it isn’t banks that bear the interest rate risk, but the taxpayers who backstop banks. How sensible is this? What about a compromise, like floating rate mortgages with floor and ceilings (for example, if you got a 4.5% floater now, its floor might be 2.5% and its ceiling might be 6.5%). That way, borrower still can make sensible budgets, since their exposure is capped, but they bear a fair bit of the risk of interest rate movements.
Point three is about the cost savings from standardization and scale, when as anyone who has dealt with a servicer can tell you, it is often at the expense of service quality.
Point four, which is a naive recitation of the canard that investors can supervise banks, is refuted by Bhide’s piece. Supervision was not done in a decentralized way; instead, the greater complexity of structured credit products led investors to rely on expert opinion (ratings agencies) and models. From a related comment by Donald MacKenzie in today’s Financial Times:
The languages of today’s complex financial markets often consist not simply of words and numbers but also of technical systems. The credit crisis has shown the importance of their powers – and limits.
Although few outsiders have heard of it, the single most important language of mortgage-backed securities and similar products is a system called Intex. It includes a computer language for defining deals’ intricate cash flow rules, a graphics-based tool for designing deals, and a truly remarkable computerised “library” of the parameters of the underlying asset pools and the cash flow rules of more than 20,000 deals….
Intex’s power as a language is to make instruments such as mortgage-backed securities mentally tractable. I confess I’ve always found them daunting. The rules governing a deal can occupy hundreds of pages of impenetrable legal prose, and the economic value of the deal’s tranches depends on three complex characteristics of the underlying mortgage pool: the rate at which borrowers prepay (redeem their mortgages early), their propensity to default, and the loss severity (the proportion of the debt that cannot be recovered if a borrower defaults).
In July a friendly banker showed me Intex in action. He chose a particular mortgage-backed security, entered its price and a figure for each of prepayment speed, default rate, and loss severity. In less than 30 seconds, back came not just the yield of the security, but the month-by-month future interest payments and principal repayments, including whether and when shortfalls and losses would be incurred. The psychological effect was striking: for the first time, I felt I could understand mortgage-backed securities.
Of course, my new-found confidence was spurious. The reliability of Intex’s output depends entirely on the validity of the user’s assumptions about prepayment, default and severity. Nevertheless, it is interesting to speculate whether some of the pre-crisis vogue for mortgage-backed securities resulted from having a system that enabled neophytes such as myself to feel they understood them.
Yves here. It may seem churlish to point fingers at Intex (“Its’ a tool! You can have operator error with any device”), but pervasive use of models allows people to think all too superficially about situations. I noticed a marked decay in the understanding of businesses when PCs became widespread. Yes, doing projections and multiple scenarios became trivial. But in the stone ages of finance, bankers and analysts had to look at financial statements and go into footnotes to find details to put into spreadsheets, and they had to do any massaging to make the presentation comparable themselves. The grappling with the data produced a far greater appreciation of what the underlying reports actually contained, and also meant any scenarios were thought about before being analyzed and any not-pretty results were given more serious consideration. Now, it’s trivial to keep tweaking a model until it tells the story needed to support a sales pitch. The degree of abstraction has made it all to easy to airbrush risk out.
To return to Bhide, his piece provides further grist for thought, and I hope you will read it in full.