GearGrind: Regulators to Study Requiring Standardized Algorithmic Computer-based Models to Pinpoint Derivatives Risks

A little noticed provision in the Dodd-Frank Wall Street Reform and Consumer Protection Act, which President Obama signed into law last week, calls for regulators to study the feasibility of requiring use of “standardized algorithmic descriptions for financial derivatives."

The provision requires a joint study by the Securities and Exchange Commission and the Commodity Futures Trading Commission that examines the viability of requiring “the derivatives industry to adopt standardized computer-readable algorithmic descriptions which may be used to describe complex and standardized financial derivatives."

"The algorithmic descriptions defined in the study shall be designed to facilitate computerized analysis of individual derivative contracts and to calculate net exposures to complex derivatives. The algorithmic descriptions shall be optimized for simultaneous use by - commercial users and traders of derivatives; derivative clearing houses, exchanges and electronic trading platforms; trade repositories and regulator investigations of market activities; and systemic risk regulators."

The provision, authored by Rep. Bill Foster (D-Ill.), a physicist and founder of a theater lighting company used by the Rolling Stones and Broadway touring shows, is nearly identical to Foster's original language which made it into the House's version of the financial reform bill, which passed Dec. 11.

When we interviewed Foster back then to try to understand more of what he or Congress was calling for in asking for such algorithmic, standard derivative models and measures, he said it was so that credit default swaps (CDSs) and other derivatives, most of which must be reported to databases to which regulators will have access (called "trade repositories" in the Dodd-Frank act), could be easily searched and analyzed by computers in an agreed-upon baseline, or standard way. The idea is that once the views of the market are centralized, there has to be an agreed-upon electronic method for measuring the risk within it. So firms can more precisely understand exactly where and at what level their exposures to derivatives augur further scrutiny or rejiggering, and the regulators can monitor more accurately the risk levels for these firms and the overall market by having the banks and other of the market's counterparties report their derivatives trades into the databases in a uniform way.

Foster confirmed today that the aforementioned conveys the gist of his provision, but in his own words added:

"Once the data of the market are centralized, there should be an agreed-upon electronic method for accessing that data so that each firm – and the regulators – can estimate the risk associated with different positions in the market."

Firms can also know better when to act to thwart trouble by knowing the risk standard to which they'll be held by regulators. The terms of many of these complex contracts can remain lost in the paper morass that has long been a congenital part of derivatives trading.

"In this bill there’s a mandate that all these complex OTC derivatives are stored in a trade repository," Foster said. "But if the trade repository consists of a big stack of legal contracts, it’s almost useless. Whereas if the trade repository consisted of a set of algorithmic descriptions the computers can rip through and say ‘hmm, this is interesting, through a series of complex derivatives, institution X was setting up a very strong position shorting institution Y.’ Then this is a very interesting question that can be reasonably asked, given the electronic form of these, by a regulator. It would be almost impossible to ask that question if you had to do it with a bunch of lawyers plowing through the individual legal clauses of a big stack of contracts. So this gives them a fighting chance of being able to plow through that."

Many custom-tailored derivatives and even a good number of contracts considered plain vanilla or generic are still executed using paper and/or the phone, which makes analyzing them effectively or efficiently impossible. Firms only recently began using email - also problematic because of its serial (one-by-one) unstructured peculiarity - to assign trades to third parties on transactions they desire to exit or take the opposite wager or position. This arcane and labyrinth nature, specifically in credit default swaps, is what precluded easy remedies during and after Lehman Brothers collapsed, and as AIG and Bear Stearns teetered toward bankruptcy and were ultimately rescued in government bailouts, when quick solutions – namely clearer views of these firms' large amounts of outstanding trading positions in derivatives – were direly needed.

The sheer volume of trades can make extremely difficult the task of tracking down all outstanding transactions to find out which of these contracts cancel (or perfectly or partially hedge) each other out - a process called "netting" - particularly when the instruments lack standardized electronic descriptions.

Also, valuing the derivatives has proven quite tough, especially when only a handful of banks - five currently - essentially control the market. Buy-side firms like hedge funds have grown to distrust both the "marks" - monetary values - the banks place on swaps these firms own, and the speed at which the banks are willing to be moved to change them, especially, these firms say, when the market moves against (financially decreases the worth of, or makes it expensive to hold) the banks' side of these transactions. It's often during times of stress or volatility when disputes erupt between counterparties over the worth of a credit default swap that one has bought from or sold to the other.

The CDS market lacks key standards and infrastructures like comprehensive trade repositories that would allow for better management of counterparty risk. There was unprecedented volatility in the credit default swaps market triggered by the bankruptcy of Lehman Brothers, the 158-year-old investment bank, and the rescue of AIG, the world's largest insurer. Besides spikes in trading of CDS linked to those firms, there were sizable increases in unwinds and novations, or trade assignments, of swaps for which Lehman or AIG were counterparties. Firms attempted en masse to cancel or reassign ("novate") derivatives contracts they had with either, both of which had been major players in the market - Lehman until Sept. 15, when it ceased to exist (but for its broker-dealer arm in the form of a swaps cleanup crew; Barclays a day later purchased the real estate of bankrupt Lehman's broker arm, and agreed to support some of its trading positions), and AIG until it began to unwind the operations of its swaps division in October 2008.

"If you look at the conversations I’ve had with the people that were trying to get something useful out of the carcass of AIG," Foster said, "they are forced to go through with their lawyers all the legal details of these very complicated derivatives contracts. And it’s a very expensive thing. And even when they’re trying to offload these assets, it’s very expensive even to get third parties to bid on them, because any third party that’s thinking of bidding on these, has to go through and have their lawyers plow through and figure out the true implications of all this and then stuff it into their valuation models to figure out what the fair value of these things are. So all of that would also be a much, much easier task if there was a standard electronic format, where you could potentially just give whoever was sitting on this big set of AIG trades, you could say ‘okay, here is the electronic description of all of these customized derivatives,’ and just send out that list of all of these derivatives contracts to all the big players who would be candidates to buy them. They could jam them into their valuation models and come up with a bid. Had this been in place at the time that we had to deal with the carcass of AIG, it would have made life much easier."

According to the act, regulators will also probe "the extent to which the algorithmic description, together with standardized and extensible legal definitions, may serve as the binding legal definition of derivative contracts. The study will examine the logistics of possible implementations of standardized algorithmic descriptions for derivatives contracts. The study shall be limited to electronic formats for exchange of derivative contract descriptions and will not contemplate disclosure of proprietary valuation models."

The SEC and CFTC are to coordinate the study with international regulators and financial institutions "as appropriate and practical," and submit a final written report of their findings by March 20 to the House agriculture and financial services committees; and the Senate agriculture and banking committees.

The CEO of an electronic platform that the big banks support in trading bonds and derivatives with buy-side firms like hedge and mutual funds said when we asked about the provision on background back in December that the bill seemed to be looking primarily to address and "demystify" the more tailored derivative contracts to make their risk components more transparent. "This particular effort appears to be more focused on the customized contracts, versus the standardized instruments like CDS indices and single-names that seem ripe for e-trading," the CEO said through a spokesperson.

Kevin McPartland, an analyst at Tabb Group, emailed us then that the provision seemed a plausible way to make the technical aspects of the derivatives market more uniform and contracts fungible, meaning interchangeable. But he said regulators should be careful to let the industry set those standards. (The "ISDA" he refers to below is the International Swaps and Derivatives Association, the main lobbying group for the market that sets derivatives standards. "FpML" refers to the Financial products Markup Language standards that make some derivatives contracts computer-readable by making the data fields included in those contracts uniform; the ISDA develops FpML standards, which are based on ISDA's template for derivatives contracts.)

"From a quick read, this feels like a lot of words to explain what is a very simple suggestion: To mandate a standard electronic format for all derivative contracts," he said. "For example, if you were to take the standard ISDA contract for a vanilla CDS and translate that into FpML or something similar, and require that all CDS transactions be stored in that format, this, in theory, would allow any trade done with any counterparty to be read and understood by any other market participant or regulators. Even for complex derivatives, all have a good handful of value [data] fields that exist no matter the complexity (such as cash flows, duration, etc.), so this is not a completely impossible idea.

"My concern however, is that Washington should not begin mandating the use of technology standards. Even in the tech industry, it is not the federal government that tells us how to send e-mails – market forces and industry groups set those rules. The same should be the case for Wall Street.

"If reporting of all OTC trades is mandated, as it likely will be, the process will require some level of automation to handle all the volume. That being the case, let the CFTC [and/or SEC] work with industry participants to agree on a standard electronic format."

Not all derivatives contracts are submitted to the Depository Trust and Clearing Corporation (DTCC), The Street's giant utility for processing and settling securities which runs a database for storing swap trades: The banks decide whether to put their OTC trades into DTCC's Deriv/SERV trade matching and confirmation system and its database - called the Trade Information Warehouse. (Deriv/SERV feeds the warehouse.) Therefore, not all swaps trades are housed or reported there. So claims that DTCC and the industry make of the various percentages of CDS or derivatives that are somehow electronically processed via DTCC do not refer to the whole swaps market; they actually refer to only the derivatives the banks consider "eligible," or decide to submit, to DTCC. So these numbers actually describe a subset of a subset of the market. It's important to remember that when reading the oft-repeated claim that DTCC processes "90 percent" of credit default swaps: That's 90 percent of the CDS the banks have decided to submit to DTCC.

MarkitSERV, a joint venture between DTCC and Markit, and the InternContinentalExchange's ICE Link, are each working on their own solutions (meant to be compatible) to make novations more automated, essentially by combining into one what used to be two steps - getting a counterparty to consent to a trade assignment as part of confirming the terms of the trade.

Regarding valuations, a credit default swaps pricing engine has been freely distributed by ISDA since early last year, after the industry was pushed to provide a method that could show some degree of objective analysis following the numerous disputes among counterparties over the worth of the contracts they held with one another. Whether the result - a bank-designed pricing engine - can truly be considered objective remains part of the debate over industry-applied standards.

Developed by JP Morgan Chase & Co., the model - now called the ISDA CDS Standard - was transferred to the trade group and made available for download in February 2009. Markit Group, a provider of credit swaps data and valuation services, acts as administrator for the engine. Markit released in March last year XML specifications for calculating standard interest rate curves based on fixed 1 percent and 5 percent coupons (interest) for generic credit default swaps that the industry decided to establish to make certain plain vanilla contracts more uniform and easier to trade or exchange.

The Street reckons that if its pricing model becomes the industry's baseline standard for valuing credit derivatives, it will limit disputes and cut down on trading and processing problems.

When asked about ISDA's standards, Foster called them "a very good starting point."

"For example," he said, "all of the legal boilerplate that they have that defines what it means to be this type of credit event - what exactly the rules are for deciding whether a credit default swap has been triggered or not: Those are key elements in the logical description of this that will remain useful. But what will be different and are different for these very customized and hard to understand credit default swaps is that there will be an algorithmic description of the logic behind them, which should allow firms to much more rapidly evaluate them, both those that are interested in making an open bid on these in the market, and secondly, for holders to understand their risk exposure."

An aside: All this talk of building more accurate derivative models reminds us of a package of stories we wrote on trading using university supercomputers and renting online retailer Amazon's spare computing power. For the former piece, we interviewed James Glimm, a distinguished professor and chair of the department of applied mathematics and statistics at Stony Brook University, based on Long Island and one of the intellectual ground zeros for quantitative finance. Glimm told us about a model Haipeng Xing, one of his colleagues in the school's computational finance department, was focused on building that could detect "change points," meaning a tool that could alert players to huge market cataclysms - rare "black swan" events or implosions - as they're first happening - when all asset values are appearing to become correlated and beginning to drop regardless of description, before the whole market plummets, or falls off a cliff.

"A change point means that you have a break with the past," Glimm said. "So an example would be a breaking bubble. In fact in finance, the major surprises are negative. The good news comes in dribs and drabs and the bad news comes on Friday afternoon. So what happens when you have some big change… is that you have all these random variables and normally they’re independent… and so you can spread your risk around – you buy a little bit of one and you buy a little bit of the other and you sort of say, ‘well if one goes up, the other will go down and so I’ve covered my risk.’ Now, when something happens, when you have a change point, everything’s correlated – you have a little bit of this, a little bit of that – but they’re all going to behave in tandem, and that’s the nature of a crisis. He has a model, and since it’s a general model, it can’t predict precisely when, [but] the idea is to have it predict when the volatility is likely to have a jump.

"You want to get your return without the risk," Glimm continued. "You want to be risk neutral but get a positive, high return. That’s sort of the use of the Black-Scholes strategy," an option pricing model that has been used to predict market direction, but that has been shown to fail to take into account the risks of large, market-moving events.

"What goes wrong is that the volatility normally fluctuates around a little bit - it doesn’t change too much - but it can have huge jumps," Glimm explained. "So it can sort of double overnight. Like if Greece defaults on their bonds, and then all of a sudden everyone goes into a panic. Volatility jumps and so it’s called a change point. So you have all of your models and they’re suddenly wrong. So [Xing’s] developing a theory which will detect the change point as it’s in the process of happening, so you don’t have to wait until you’re broke to discover your portfolio is worthless; while there’s still a chance to get in there, he will have a model that says ‘hey, something’s happened, we better go to plan B.’ One might spot it as it’s happening with these theories so that you could at least have a bit more of a rapid response."

All of which we were reminded of this Sunday while laying about reading the "Economics focus" section of the latest issue of The Economist, which describes central bank economists, policy experts and computer scientists putting their heads together to build unconventional "agent-based models" that could detect early the next financial crisis. These models differ in that agent-based models make no assumptions that markets are efficient or always adhere to a narrow band of "normality" or "equilibrium," whereas conventional models do: The last crisis pointed out the serious flaws in said presumptions, which are held by what's called the "efficient market hypothesis," which basically makes no room in its calculations for steep downturns. Essentially, conventional models can't account for crashes because they don't see them. They can't because they aren't made to look for them. Whereas agent-based models replicate reality better by sending out "agents" ("simulations" for simplicity) that independently interact with one another and whose reactions can be proportionally larger than the cause that triggered them, just as they are in real life panics or in the "herding" behavior exhibited by investors under stress, or in responding to some market bias that, like a giant Russian doll, can be built up and gain in popularity or volume to become conventional wisdom that falls over and rolls in one destructive direction, flattening everything. The Economist says Bank of England's Sujit Kapadia is trying to model the complex derivatives market and the interconnectivity of its participants to shore up the weak links in this web.

All these calls for better models seem to us to naturally stimulate rewarding business opportunities for anyone smart and innovative enough to tinker toward real solutions. So whiz-bang: Go get at it!

One of the most interesting responses to Foster's call for assessing ways to standardize methods of determining derivatives exposure came from Dr. Andreas Binder, managing director at MathConsult, a Linz, Austria-based firm that develops and sells derivatives pricing and risk analysis software. Binder pointed out the challenges in finding a baseline tool for assessing exposures to instruments as complex as derivatives, particularly if the goal is to uncover the potentially risky correlations between all of them. Doing so is complicated by the fact that just a single derivative includes multiple variables that can be impacted by a number of fluctuating circumstances, such as changing interest rates or a company's ability to pay down debt, all of which needs to be understood, or modeled, accurately.

A derivative may itself be comprised of several other derivatives, the terms of which can be customized, like in the case of a "swaption," which is a swap on an option, or a "callable inverse snowball," an exotic instrument with two funding legs, or payment streams, which requires computation of past coupon (interest) rates with whatever the current, floating rates are.

Binder's mention below of a "copula trap" refers to the conundrum that measuring complex instruments accurately requires inclusion of myriad variables, yet because one must control for the volume of those variables to make the measurement practical, doing so may preclude the assessment of too many interactions, or precisely those interactions needing to be examined for the analysis to be close to accurate, while including too many could result in "noisy," or inaccurate, results as well. Many analysts, in fact, blame The Street's nearly wholesale adoption of what's known as the "Gaussian copula function" for creating the credit crisis, because it assessed default risk in mortgages by using the prices of credit default swaps which had only been around in number since the housing boom began its rise, and therefore were wildly inaccurate measures of that market's risk.

"In market risk, there is some standardization by exchanges or – in the case of OTC swaps – by ISDA. However, the standardization of swaps does not mean that the term sheets under such standardized schemes are easy to value and that their risks are easy to assess.

“In the credit market, there is some standardization for single name CDSs and probably also for CDOs [collateralized debt obligations; see definition below], but there is no central evidence on the nested papers of papers of papers lying in special vehicles.

“As I understand, the Dodd-Frank Act aims to install such a central database. If this database was installed, I am not completely convinced that a) you would obtain a transparent view on the market, and b) that any fantastic supercomputer would be able to tell you all the nested positions of market participants in such papers.

“Mathematically speaking, the difficulty of [measuring risks of] multi-name, credit-linked instruments lies in correlation," Binder explained. "For assessing the various influences of derivative positions, this magic computer has somehow to calculate numbers by assuming a model for the various influences [on derivatives' prices and risk exposures]. Depending on the assumptions of such a model, it will deliver different numbers. So, in order to avoid the copula trap, almost everything would have to be forbidden, which, from my point of view, is no solution."

We sent queries to a bevy of derivatives vendors - including Andrew Kalotay, FinCAD, Markit, Numerix, Pricing Partners, Quantifi Solutions, SciComp, SuperDerivatives, Icap's Trioptima and MathConsult - to see if they sniffed any business opportunities that might emanate from the regulators' feasibility study on creating a standard derivatives description and baseline measure of net derivatives exposure. So far, besides MathConsult, just Andrew Kalotay, Quantifi and Superderivatives emailed us back, although Superderivatives has yet to provide responses after saying it would.

Such efforts reside "outside our core domain," Kalotay said through a spokesman.

Quantifi said through a spokeswoman that market forces were already taking care of developing that which might be required from the results of the study. Such work, she said, includes the industry upgrading FpML, the adopted standards that make some derivatives contracts computer-readable by making the data fields included in those contracts uniform. The industry work, she said, also includes the recent proposal from the SEC that would require the conversion into electronic descriptions any paper-based details of the terms of loan payment streams for which bonds like mortgage-backed securities collect and are made up of, so each payment stream specification, or contractual provisions on said cash flows, would be downloadable into Python, a computer-readable programming language. The "waterfall" she refers to below is the structure determining which bond investors are paid first, based on receipt of loan payments that make up the bond.

"I’m not sure we have much to say on this," said the Quantifi spokeswoman. "The reason being is that the market is already looking at a solution for algorithmic description (which is different from algorithmic valuation – the bill is focused on description). There’s FpML obviously and then there is the recently proposed Python-based waterfall descriptions for collateralized contracts. Since the market is already looking at this and may choose to operate more electronically even without government intervention, Quantifi is participating in industry-wide efforts but isn’t necessarily pursuing this as a dedicated 'business opportunity,' but instead, treating it as an industry development as it would anything else."

Definitions:

A collateralized debt obligation, or "CDO," is a type of asset-backed security, or "ABS." An ABS is a pool of individual loans that can include mortgages or auto loans or credit card debt, which is packaged into a bond and sold to investors. An ABS comprised of mortgages is often called a mortgage-backed security, or "MBS." A CDO pools together any of these or other types of debt, loans or other bonds, the tranches, or pieces of which have different risks and maturities associated with them. Think of it as an "ABS of an ABS."

Credit default swaps are often described as being like insurance, but they are also instruments enabling firms to bet against, or "short" particular debts, when the belief is that a certain bond or debt is overvalued or headed for default. Some firms use them to hedge bonds they own against default, others to bet the bonds' value will sink. One need not own the bonds to which the swaps are linked to short, or bet against them. Critics say the latter, often referred to as "naked CDS," is comparable to buying insurance on another person's property or life, which is illegal due to the perverse incentives it can create in those so inclined to seek the destruction of said person's house or life, for money. Credit default swaps are insurance or shorts linked to the value of bonds issued by corporations; governments, from nations to municipalities; mortgage lenders; credit card issuers; auto lenders. Conceivably, CDS can be linked to any bonds or debt. With CDS, one often hears about the "spread," which represents the risk of the debtor defaulting on its obligations to pay back what it owes or what it was lent. That risk is measured or made tangible by being converted into a fee: The spread determines the annual fees, much like an insurance premium, a swap buyer must pay to the seller, which functions much like an insurer. The deal requires the seller to make a lump payout to the buyer if the debtor goes bankrupt, or if the debtor fails to pay on time what it owes. (Or in the case of "pay-as-you-go" mortgage CDS, the seller must make incremental payouts on each loan on which a homeowner defaults or fails to pay.) It also requires the seller to post collateral (make payments) to the buyer if the risk of default of the bonds underlying the CDS rises. The greater the risk grows of default on the bonds linked to the CDS, the more the CDS is worth, to the buyer betting against or "insuring" the debt; and the more costly it becomes for the seller. Vice versa if the debt's default risk decreases: The CDS becomes more valuable for the seller as the debt "being insured" or shorted is deemed safer, which ramps up the cost of "insurance" or "the short" for the buyer. Yet, buyers of CDS can potentially make money several ways: By selling CDS after their value rises to parties needing to cover their exposure to the troubled debtor involved, by receiving collateral from the CDS seller as the risk of default on the underlying debt rises, and/or receiving from the CDS seller a lump or incremental payout (like an insurance settlement) to cover the default of the bonds if they totally tank. The value of those payouts sellers must make to buyers - the insured or shorted amount - is often in quantities much larger than the sum of fees the buyer pays to the seller to hedge (insure) or short (bet against) the debt the swap references. However, most speculators cash out by selling CDS to others needing the protection, because waiting for defaults poses counterparty risk: that the seller could at any time become financially unable to make the lump or incremental payment or post collateral.