Sign Up

The Financial Crisis: An Accident Waiting to Happen

Did Wall Street’s over-reliance on mathematical models undermine the sound judgment of risks?

June 8, 2009

Did Wall Street's over-reliance on mathematical models undermine the sound judgment of risks?

President Clinton, whose administration midwifed the first large-scale production of financial toxins, blames the current crisis primarily on the absence of good investment opportunities besides housing in the Bush administration.

Others have indicted the Federal Reserve’s monetary policy, the rating agencies, and even the S.E.C. for abolishing the uptick rule, which discouraged short selling.

In reality, elected officials and appointees from both political parties — and respected economists — had so undermined the banking system that anything could have triggered a collapse.

According to the prevailing wisdom, the crisis was the result of a regulatory apparatus that had fallen behind the development of modern financial theory and practice. There were too many gaps in the regulation, and these need to be filled.

But there is little recognition of the role that regulation actually played in fostering the crisis. Thanks to the regulations, the wizards of Wall Street were allowed to lever up their balance sheets — and their bonuses — to levels far beyond what private lenders would tolerate.

Without F.D.I.C. insurance, for example, banks that engaged in highly-levered speculation — or that extended credit to investment banks and hedge funds that engaged in such speculation — would have faced great difficulty in attracting the deposits they needed.

The conventional wisdom is also defensive: It holds that by filling the right regulatory gaps, the financial status quo can be saved from its excesses. But is the new financial technology really worth saving? How do the securitization of the credit that was previously extended by banks contribute to economic prosperity? And how do the derivative instruments and trading based on this securitized debt do so?

In 1987, Lowell Bryan, a McKinsey & Company director, wrote that “a new technology for lending-securitized credit has suddenly appeared on the scene. This new technology has the capacity to transform the fundamentals of banking, which have been essentially unchanged since their origins in medieval Europe.”

Bryan predicted that traditional lending might soon become obsolete: “About half of all debt in the national economy is raised through securities; that number might increase to 80% in the next decade.” The new technology, he argued, offered more checks and balances than traditional banking:

“Under a securitized credit system, in which an outside agency assigns a rating to the issue, credit risk will likely be properly underwritten before investors will buy an issue. In many cases, another third-party credit underwriter (a bank, a finance company, or an insurance company) must
guarantee a portion of the credit risk in the issue. So at least one and often two skeptical outside parties review the credit underwriting before the issue can be placed with investors.”

In contrast, he suggested, “loan risks depend entirely on self-definition by the institution making the loan.” Similarly, the rates of securities were set by an objective market, not by the subjective judgments of bankers. And securitized debt was “attractive to individuals, pension funds, and other investors who either can’t or won’t assess credit risk and would rather let rating agencies do the job.”

The claim that the automated processing of hard information provided by a centralized source is usually a superior substitute for the subjective judgments of a banker — F. A. Hayek’s “man on the spot” — ignores the unquantifiable uncertainty that is an important feature even of seemingly routine lending decisions.

Using a credit score produced by feeding a few items of hard data into a mathematical model to assess the likelihood of default assumes that all risks are quantifiable. And that’s just one of the many assumptions at work.

For instance, credit-scoring formulae also assume that the probability that all loans of a certain kind will default derives from exactly the same risk factors. These risk factors are all combined or “weighted” in exactly the same way, and somehow, an omniscient modeler knows the right weighting scheme.

Such assumptions would be unacceptable in other walks of life: replacing “routine” felony trials with a scoring model is inconceivable, whatever the cost savings might be. Nor do economics departments economize on the costs of hiring even entry-level faculty or Ph.D. students by using predictive quantitative models (save in countries like France, where faculty are hired on the basis of objective scores in a competitive national exam).

Like criminal trials and faculty hiring decisions, the traditional lending process implicitly took into account unquantifiable uncertainties and the uniqueness of individual circumstances. The difference between the information produced by a loan officer’s visits and that offered by Dun and Bradstreet is just a matter of soft rather than hard.

Visits produce information that is wider in its range (and can cover private information that is not available to Dun and Bradstreet) and better tuned to the specific circumstances of the borrower. For instance, a commercial loan officer may take note of changes in the number of cars in the visitors’ lot of an industrial distributor, but ignore such changes for an Internet retailer.

Similarly, loan officers and committees traditionally used a wide range of information (including both quantitative data on past and projected financial performance and qualitative observations about competitors and customers) to construct a coherent “case” or “narrative” — rather than plugging data into a formula.

This may have amounted to overkill in certain kinds of lending: mortgages with high down payments in stable housing markets, for instance. But it is hard to imagine that mechanistic lending is an appropriate rule for most credit decisions, and that case-by-case ought to be reserved just for unusual situations.

Similarly, embedding financial transactions in long-term relationships instead of conducting them at arm’s length in an “objective” marketplace has merit in many seemingly mundane contexts, and not just for “the most complicated, innovative or risky financial transactions.”

Banks whose lending far exceeded their base of long-term depositors have discovered that it is dangerous to rely on funding by fickle strangers in wholesale money markets. A similar situation exists with the extension of credit. A financial institution that underwrites securitized credit for resale becomes, to a significant degree, a sales agent for the borrower.

Of course, sensible sales agents who value their relationships with customers will exercise some care in what they sell — nonetheless, the degree of care is diluted by the expectation that customers will do their own analysis, and by the absence of any direct financial risk to the sales agent.

Thus, an underwriter of debt cannot be expected to exercise the prudence of a banker making a loan that will remain on the bank’s balance sheet.

Long-term relationships between lenders and borrowers have great value even after credit has been extended, akin to the benefits of shareholder-manager relationships. Borrowers can share private information with lenders just as corporate “insiders” could (if not barred by law), and thus have a greater opportunity to send early warnings of danger.

In addition to self-interested restraints on opportunistic behavior, because the parties know they are stuck with each other — a banker cannot dump a thirty-year loan as easily as a mutual fund can sell a bond — there may develop an additional sense of mutual solidarity. A banker may thus renew a line of credit in hard times where an arm’s-length purchaser would not roll over the same issuer’s maturing commercial paper.

Renegotiating the terms of a loan with one banker is easier that corralling many dispersed bondholders to discuss the modification of bond covenants. One of the consequences of the slicing and dicing of mortgage loans is that it is now often practically impossible for homeowners in default to work things out with their lenders, as they might if their mortgage had a single owner, especially one located at the nearby branch of their bank.

Why was there such a mass displacement of long-term, relationship- and judgment-based lending by arms-length securitization? In the narrative offered by Rajan and several other economists, exogenous technologies played a deterministic role, inexorably forcing changes in regulation and financing arrangements. But technology might, instead, have facilitated relationship banking.

For instance, collaborative software (such as Lotus notes) could have improved the capacity of large lending teams serving far-flung borrowers to share a wide range of data, observations, and judgments. The outcome was not predetermined.

In fact, in the story that I have told here, the increased share of securitized financial assets was driven mainly by the beliefs of financial economists and regulators.

Economics has underpinned securitization through its embrace of mathematical models to the exclusion of other perspectives — and through a complementary tendency to ignore the downside of liquidity and arms-length relationships.

Regulation has brought this way of thinking into practice in two paradoxically related streams: by increasing the scope and effectiveness of the New Deal securities acts and subsequent rules that fostered the growth of arms-length transactions in corporate control; and the progressive dilution of New Deal banking acts, which nurtured and protected long term relationships.

This is the complicated story that may explain why developments in mortgage banking, of all things — traditionally the plodding, conservative bread-and-butter of depository banking — should have led to the implosion of the world economy.

This feature is excerpted from an article to appear in a special issue of “Critical Review” on the causes of the financial crisis, which can be ordered here.

Takeaways

Entering a few items of hard data into a mathematical model to assess the likelihood of default assumes that all risks are quantifiable.

Banks whose lending far exceeded their base of long-term depositors have discovered that it is dangerous to rely on funding by fickle strangers in wholesale money markets.

Similar assumptions of risk would be unacceptable in other walks of life: replacing "routine" felony trials with a scoring model is inconceivable, whatever the cost savings might be.

The conventional wisdom is also defensive: It holds that by filling the right regulatory gaps, the financial status quo can be saved from its excesses. But is the new financial technology really worth saving?