After the Crash: How Software Models Doomed the Markets
The above also is the title of a November 21st 2008 editorial by The Editors of Scientific American. The editorial's sub-title is "Overreliance on financial software crafted by physics and math PhDs helped to precipitate the Wall Street collapse". This editorial is well worth reading both today and in the months and years to come, as all parties consider the form and regulation of global financial markets after we get through the current credit crisis.
The editorial begins:
If Hollywood makes a movie about the worst financial crisis since the Great Depression, a basement room in a government building in Washington will serve as the setting for a key scene. There investment bankers from the largest institutions pleaded successfully with Securities and Exchange Commission (SEC) officials during a short meeting in 2004 to lift a rule specifying debt limits and capital reserves needed for a rainy day. This decision, a real event described in the New York Times, freed billions to invest in complex mortgage-backed securities and derivatives that helped to bring about the financial meltdown in September.
In the script, the next scene will be the one in which number-savvy specialists that Wall Street has come to know as quants consult with their superiors about implementing the regulatory change. These lapsed physicists and mathematical virtuosos were the ones who both invented these oblique securities and created software models that supposedly measured the risk a firm would incur by holding them in its portfolio. Without the formal requirement to maintain debt ceilings and capital reserves, the commission had freed these firms to police themselves using risk tools crafted by cadres of quants.
The staff at Toomre Capital Markets LLC has long admired this publication, partly since both Lars and Aldon started in technical fields before moving to Wall Street in the 1980s. Immediately before starting at Lehman Brothers, Lars Toomre was at M.I.T. and Aldon Hynes was at what then was one of the Mecca's of industry research, Bell Labs.
Where did we work in Lehman Brothers? We each started work in the mortgage department focusing on the very software that let Lehman Brothers and other investment banks slice and dice pools of assets into various classes (or tranches) of debt securities then known as Collateralized Mortgage Obligations ("CMOs") and now as Collateralized Debt Obligations ("CDOs"). We were two of the first quants hired by Lehman Brothers' fixed-income division working on some of the early software that this editorial targets!
The editorial continues:
The software models in question estimate the level of financial risk of a portfolio for a set period at a certain confidence level. As Benoit Mandelbrot, the fractal pioneer who is a longtime critic of mainstream financial theory, wrote in Scientific American in 1999, established modeling techniques presume falsely that radically large market shifts are unlikely and that all price changes are statistically independent; today’s fluctuations have nothing to do with tomorrow’s—and one bank’s portfolio is unrelated to the next’s. Here is where reality and rocket science diverge. Try Googling “financial meltdown,” “contagion” and “2008,” a search that reveals just how wrongheaded these assumptions were.
This modern-day tragedy could be framed not only as a major motion picture but also as a train wreck or plane crash. In aviation, controlled flight into terrain describes the actions of a pilot who, through inattention or incompetence, directs a well-functioning airplane into the side of a mountain. Wall Street’s version stems from the SEC’s decision to allow overreliance on risk software in the middle of a historic housing bubble. The heady environment permitted traders to enter overoptimistic assumptions and faulty data into their models, jiggering the software to avoid setting off alarm bells.
The causes of this fiasco are multifold—the Federal Reserve’s easy-money policy played a big role—but the rocket scientists and geeks also bear their share of the blame. After the crash, the quants and traders they serve need to accept the necessity for a total makeover. The government bailout has already left the U.S. Treasury and Federal Reserve with extraordinary powers. The regulators must ensure that the many lessons of this debacle are not forgotten by the institutions that trade these securities. One important take-home message: capital safety nets (now restored) should never be slashed again, even if a crisis is not looming.
For its part, the quant community needs to undertake a search for better models—perhaps seeking help from behavioral economics, which studies irrationality of investors’ decision making, and from virtual market tools that use “intelligent agents” to mimic more faithfully the ups and downs of the activities of buyers and sellers. These number wizards and their superiors need to study lessons that were never learned during previous market smashups involving intricate financial engineering: risk management models should serve only as aids not substitutes for the critical human factor. Like an airplane, financial models can never be allowed to fly solo.
Toomre Capital Markets LLC ("TCM") very much agrees that risk management models should serve only as aids and not as substitutes for the critical human factor called judgement. TCM has previously written about quants, notably including this post Businessweek: Math Will Rock Your World. All of these "new" mathematical models for both new businesses and finance might be right for all of the observable data. However, as quants themselves, both Lars and Aldon continually remind people, particularly non-technical types, that they need to know and understand the assumptions that went into the construction of that particular model.
Take, for instance, that original CMO model at Lehman Brothers. Over the years that "simple" cash flow model morphed from first being able to only handle sequential pay bonds into one that allowed for both fixed and floating rate collateral and debt structures, weird prioritization of both principal and interest cash flows, the inclusion of possible subordination and the possibility of reserve funds (either funded up-front or from the excess spread in the securitization during its first few years).
In the early years, one constant prepayment rate was used to value the resulting tranches. However, as dealers and investors gained a better understanding of homeowner behavior, the need for variable month-by-month gained importance. A later "improvement" demanded by leading market participants was prepayment forecast modeling which in turn was fed into yet another improvement, a model that calculated the Option Adjusted Spread ("OAS").
An OAS value attempts to express the spread over risk-free interest rates that would result if the "optionality" embedded in a security were either bought back or sold at the then current option prices in the open market. In the case of a mortgage pool, the security holder is generally short the option given to the homeowner to prepay the underlying mortgage at any time. And generally there is a fair amount of prepayment activity in all mortgage pools as families move for job reasons, split up, retire or relocate. OAS is a "short-cut" way of trying to estimate that prepayment risk.
Returning back to that original "simple" CMO model, one gains a small sense of the many assumptions that went into the calculation of the OAS value for a particular CMO tranche. Of course, the value generated by the Lehman Brothers model often varied, sometimes slightly and sometimes greatly, from those generated by those models developed and used by Goldman Sachs, Morgan Stanley et al.
Often the difference between these models was in one of the assumptions. One firm might use the three-month moving average of historical price data to determine its volatility estimate assumption. Another might use the 30-day average. However, the "average" institutional investor often did not want to dig into even this key assumption. Often an investment decision was made because the reported OAS value of bond A was higher than that generated by a different model on bond B.
Toomre Capital Markets LLC hopes this post reminds everyone that there are considerable dangers in rote reliance on the numbers produced by the computer. One needs to understand the assumptions that went into that particular model and understand where those assumptions are likely to break down. One also needs to use judgment that the model was coded as described and that "bad" data was not feed into the model or spreadsheet.