structured securities default, and the loss severity (the more nearly all took a fatal shortcut. proportion of the debt that cannot be Instead of analysing CDOs from the Donald recovered if a borrower defaults). bottom (the underlying pools of In July a friendly banker showed me mortgages) up, they shifted to a different MacKenzie Intex in action. He chose a particular mathematical language, which treated a INSIGHT mortgage-backed security, entered its CDO’s components (mortgage-backed price and a figure for each of prepayment securities and tranches of other CDOs), speed, default rate, and loss severity. In in effect, as if they were corporate bonds, “The limits of my language,” wrote the less than 30 seconds, back came not just with their properties inferred from their philosopher Ludwig Wittgenstein, “mean the yield of the security, but the ratings. This often led to serious the limits of my world.” month-by-month future interest underestimation, especially by rating The languages of today’s complex payments and principal repayments, agencies, of correlation among these financial markets often consist not including whether and when shortfalls components. simply of words and numbers but also of and losses would be incurred. The The one bank I’ve found that did technical systems. The credit crisis has psychological effect was striking: for the analyse CDOs based on mortgages from shown the importance of their powers – first time, I felt I could understand the bottom up was Goldman Sachs. It and limits. mortgage-backed securities. developed its own analytical techniques, Although few outsiders have heard of it, Of course, my new-found confidence and used a large “computer farm” in the single most important language of was spurious. The reliability of Intex’s New Jersey to spread the analysis over mortgage-backed securities and similar output depends entirely on the validity of multiple products is a system called Intex. It the user’s assumptions about machines, so includes a computer language for prepayment, default and severity. keeping the The limits of the defining deals’ Nevertheless, it is interesting to time each run intricate cash speculate whether some of the pre-crisis took tolerable. language came flow rules, a Intex is not cheap vogue for mortgage-backed securities Goldman’s when mortgage- graphics-based resulted from having a system that Abacus CDOs – backed securities tool for – costing one enabled neophytes such as myself to feel one of which designing they understood them. Certainly, like any was the were repackaged bank about $1.5m deals, and a language, Intex aided communication. If flashpoint of into collateralised truly a year – but it is you were planning a mortgage-backed the SEC’s debt obligations remarkable essential for deal, you could construct an Intex file, recent computerised make it available to potential investors, investigations “library” of the participants in and use it to discuss the deal’s features, – were apparently analysed this way. parameters of structured modify those features, and gauge A bottom-up analysis was – and the underlying investors’ interest. still is – expensive. It requires clever asset pools and securities The limits of the language came when quantitative analysts, multiple the cash flow mortgage-backed securities were computers and software developers rules of more repackaged into collateralised debt able to “parallelise” a program so it runs than 20,000 deals. Intex is not cheap – one obligations (CDOs), complex debt efficiently on many machines at once. user told me his bank pays about $1.5m a securities based on pools of other assets. Nevertheless, if going beyond the limits year for it – and it has competitors such You could still run Intex, first for each of of existing languages in this way helped as Bloomberg, but it is essential for all the securities and then for the CDO, but Goldman take the crucial late-2006 serious participants in structured it could be a slow process. Often, CDOs decision to liquidate or hedge its securities. included not just mortgage-backed positions in mortgage-backed securities Intex’s power as a language is to make securities, but tranches of other CDOs, (enabling it to survive the crisis almost instruments such as mortgage-backed each maybe incorporating further CDOs. unscathed), it was well worth it. I hope securities mentally tractable. I confess This multiplied enormously the number its competitors have learned the lesson: I’ve always found them daunting. of underlying mortgage pools, causing a a limited language means a dangerously The rules governing a deal can occupy single valuation run to take hours. (On limited world. hundreds of pages of impenetrable legal occasion, each of a pair of CDOs would prose, and the economic value of the buy a tranche of the other, creating a Professor Donald MacKenzie teaches deal’s tranches depends on three complex “loop” that slowed analysis). Sometimes, sociology at The University of Edinburgh characteristics of the underlying users did little more than one run using mortgage pool: the rate at which the prepayment, default and severity borrowers prepay (redeem their rates judged most likely. Those (such as mortgages early), their propensity to the rating agencies) that needed to do