The complex life

The popularity of complex derivatives products has created a need for independent pricing and data services. With users demanding ever more speed and accuracy in the delivery of pricing data, a variety of new platforms has sprung up. Clive Davidson reports

Investors are trading an increasingly varied and extensive set of securities in their search for yield. With the developments in financial engineering, new markets such as credit derivatives growing rapidly, and investors increasingly willing to take leverage to bolster returns, the use of complex instruments has surged. The challenge is obtaining independent, valid price information in these markets. At the same time, regularly valuing portfolios containing less liquid instruments can present considerable difficulties.

Traditional data sources have not typically covered complex instruments such as synthetic collateralised debt obligations (CDOs). However, established data vendors and new entrants are now developing fresh sources of information to provide customers with data on these securities. Meanwhile, those providing portfolio valuation services are extending their range, often in alliance with specialist pricing companies.

"In the over-the-counter markets, the mode in which you gather data for pricing and valuing securities and contracts requires new business models," says Michael Liberman, head of quantitative strategies and chief technology officer at BlueMountain Capital Management, a credit-focused hedge fund with $2.8 billion under management.

BlueMountain trades credit default swaps (CDSs), plus CDS indexes and index tranches, and has around 1,500 names in its portfolios. "For each credit there are different forms in which it is traded. You have different seniority, and you have different International Swaps and Derivatives Association conventions as to what constitutes a default event - for example, with restructuring or without restructuring - which imply different pricing. This all has a multiplying effect on the number of data points you need to handle those credits," says Liberman.

When a market is predominantly OTC, participants face the problem of where to get information in order to price securities or value their portfolios. In a relatively new market such as credit derivatives, this is particularly acute. However, several firms have sprung up to cater to this need. In 2003, London-based asset valuation and price verification company Markit introduced a credit pricing service based on data contributed by a number of leading banks, including Bank of America, Deutsche Bank and Goldman Sachs. The company now has 60 contributors and 400 clients, and publishes end-of-day composite prices and spreads for 2,400 credit entities.

A similar service, Valuspread, originally created by London-based Lombard Risk Systems and now operated by New York-based Fitch Ratings, covers 2,500 entities, with historic data going back to 1999.

Producing the prices for these services is a considerable task, involving the gathering of over 1 million pieces of data from global contributors, checking the information, and calculating the composites and other outputs. The checking - known as data cleaning or scrubbing - has to be rigorous, says Penny Davenport, director at Markit. On average, the company rejects approximately 45% of the CDS data it receives because it judges the prices to be stale, outside of an acceptable range, or undifferentiated (the 'flat curve' test where all spreads along a price curve are the same).

Given the volume of data and the time constraints, the process must be "highly automated and super fast", says Davenport. New York contributors, the last in the sequence after Asian and European institutions, must get their data to Markit by 2am London time (9pm New York), for processing to be completed before European markets open. Markit has employees in Vancouver, in North America's western time zone, that monitor the data and alert London if there are problems.

The pressure to improve the data is relentless, however, especially for liquid instruments. "Standards in the industry are high - firms demand excellent data quality," says Davenport. To this end, Markit recently began incorporating broker prices into the calculation of its intra-day composites. "We give a broker price a very heavy weighting in the composite to accurately reflect the fact that the credit has traded at that price. That's what traders expect," says Davenport.

Quote inclusion

Another thing traders expect, especially in the price curves for liquid instruments, is the inclusion of quotes they have seen during the day. "A bank or hedge fund will look at an end-of-day curve, and if they have seen a quote for that name that day, which is somewhat different from the end-of-day composite, they will then raise a lot of questions," she adds.

But although this model of collecting, cleaning and weighting data from contributors to calculate end-of-day composites has been widely accepted - a number of analytics, trading and risk management systems vendors have integrated Markit's data into their products, including valuation services from New York-based Moody's KMV and Massachusetts-based FT Interactive Data - the process has limitations, and other vendors are taking a different approach.

The first issue is timeliness - although it is end-of-day data, it is only available the following morning. In a step towards addressing this, Markit recently introduced a service called Markit Sameday, where it publishes separate prices for each region at the end of their trading days.

The second issue is that the aggregation and averaging of dealer quotes can lead to distortions under certain circumstances. Take the case of where a series of related contracts are traded in the market, such as a credit with different seniorities and default events that may or may not involve restructuring. Generally, the price of only one of these is quoted, with other prices derived by algorithms. Say 12 dealers contribute quotes for the main credit curve, but only four contribute quotes for the derived prices. However, these four happen to be at the high end of the main price curve. When the averaging process is applied to the derived data, it will create curves that are overall higher in the price spectrum than the main curve. "Because of how the contributions are made, you sometimes get results that don't make sense," says Liberman.

James Rieger, vice-president of global pricing services at New York-based Standard & Poor's (S&P), says questions can be asked of dealer-contributed prices. One is when were the quotes generated - market events can affect the value of an asset over the course of the trading day. Another question is who at the dealer generated the quotes? "If it was the trading desk, it would be more meaningful than if it was the sales desk. Or it might be that it was an analyst, or a mathematical model," says Rieger.

The impact of these factors can be compounded as the prices are averaged, potentially taking the calculated price further away from actual trading prices, he says. Although S&P does use dealer quotes in its valuation services, it gives precedence to actual trade data and bid/offer prices. "There may be a really good quote within the group of quotes you're averaging that truly represents the value of that asset, but if you average dealer quotes, the result may not be representative of where an institution could sell that asset in a market-place," says Rieger.

In April, S&P teamed up with Connecticut-based valuation specialist Complex Security Valuations (CSV) to extend its valuation service to cover CDSs, as well as CDOs, collateralised loan obligations and other complex and/or illiquid products. CSV uses sophisticated modelling tools and in-house expertise to form opinions on prices for S&P's customers.

An alternative source of price data is the constant stream of quotes that trading desks and buy-side institutions receive from banks and brokers via email. Although email, including closed community services such as that of financial information and analytics provider Bloomberg, offers a highly flexible, one-to-one medium that is ideal for the relationship-based nature of OTC trading, it poses a challenge for an organisation that wants to capture the market information within the messages and use this in a more formal way.

"There's a meaningful amount of differential pricing in this industry and that is where relationships come in," says Laurent Paulhac, chief executive of London-based credit data vendor Credit Market Analysis (CMA). "Email allows you to maintain those relationships, but the challenge is how to absorb this information to make quality decisions."

The problem is both the quantity of information institutions receive - a trading desk can receive 10,000 or more emails a day - and the fact that there is no standardised way of presenting quote data, either in the message format, or in the instrument or entity codes and identifiers that are used. Also, the format can change as people come and go on trading desks. "One dealer might type the price information as a string of tickers, while another might come along and present it as a table," says Paulhac.

To overcome these problems, CMA has developed artificial intelligence-based software that scans email, captures the pricing information, standardises it and stores it in a database. The software parses the informal text, recognising or deducing which is the relevant data, and extracts it. Once the information is in the database, institutions can organise and analyse it for trading or valuation purposes. Called QuoteVision, the application is now used by more than 75 institutions and is on trial at more than 30 more, says Paulhac.

Liberman of BlueMountain, which uses both Markit's CDS data service and QuoteVision, says: "For us, the most important data source is Bloomberg, where dealers send quotes through mail messages. QuoteVision processes all the quotes we get, parses them and puts them into a database in an easy-to-consume form so we can readily get access to all that data."

However, like most automated attempts to understand the vagaries of human communication, QuoteVision cannot provide complete interpretation. "The credit market is a very complex and dynamic environment, with a lot of creativity and change," says Paulhac. For the moment, software is able to automatically recognise 90-92% of email quotes. To deal with the exceptions, CMA operates a human editorial service. When QuoteVision encounters an email it cannot parse, it extracts the relevant section of text - and only that section - and sends it without identifiers as to the originator or receiver (to protect privacy) to CMA's human editors, who are based in London. If the editors are able to make sense of the text, they send their interpretation back to the client; if not, the text is deleted. In addition, to keep improving QuoteVision's automatic abilities, clients can send new formats and identifiers to CMA, which creates new rules for recognising the formats and identifiers, and these are uploaded to client installations of QuoteVision. This can be done on a daily basis, says Paulhac.

Markit is currently testing a similar application with a number of clients. The software, which goes under the working title of Markit Quotes, sits on a trading desk and scans all incoming emails to identify pricing information for credit entities or indexes, says Davenport. "It uses Markit's Reference Entity Database to identify entities, and applies sophisticated logic to make sure it's adding apples to apples to come up with the best bid and offer prices and average price for the day," she says. And these prices will be constantly updated as new quotes arrive during the day. Markit expects to launch the product formally by the end of the year, although Davenport does not expect it to differ much from the version currently under trial.

Partly to address the timeliness issue of CDS pricing services such as Markit and Valuespread (which also publishes its global end-of-day prices the following morning European time), but also to offer a source of data with different characteristics, CMA created a consortium of 26 buy-side companies, including hedge funds, large asset managers and buy-side desks at investment banks, to contribute to its alternative pricing service. "As well as the information they get via email, our contributors have their own models and views of where the market should be at any given point, and they contribute these estimates for CDSs, indexes and tranches," says Paulhac.

CMA renders the data anonymous, and aggregates and averages it to produce a consensus-based price validation service called DataVision, delivered at 5pm in London and 5pm New York time for their respective regions, "so organisations can start marking to market their positions before they go home", says Paulhac. He also claims that because DataVision's prices come from the front offices of its contributors, its subscribers are less likely to experience conflicts between front and middle offices, unlike when their middle offices are getting their prices from services that rely on contributions from back-office systems.

With data and valuation services for CDSs well under development, service providers are pushing on into other areas where there are market and regulatory demands for more frequent pricing of illiquid OTC securities and contracts. Fitch Ratings, FT Interactive Data, Markit and S&P all recently launched evaluation services for products such as European asset-backed and mortgage-backed securities and synthetic CDOs.

"A lot of hedge funds don't want to get CDO prices only from their banks, which is why we have come out with a valuation service," says Stephan Sanner, director for Valuspread in the UK for Fitch. Fitch's new service, launched in July and called Risk Analytics Platform for Credit Analytics (Rap CD), uses CDS and index price data from Valuspread and correlation data from London-based broker GFI. "Correlation data is an important input for CDO valuations," says Sanner. "However, banks don't want to give it away because it is competitive information. GFI is the only broker that sells it at the moment."

Rap CD is built on the Algo Risk market risk analytics and stress-testing technology from Toronto-based Algorithmics, which Fitch acquired last year, and uses valuation models from Algorithmics, as well as those Fitch acquired when it bought the credit analytics business of London-based consultancy Reoch Credit in July. "However, Rap CD is agnostic as to the market data and models it uses," says Simon Greaves, senior director for the product at Fitch. The company plans to source additional data and models from other providers. "In terms of models, we only want to use those that are accepted as standards in the market, and we will be transparent as to their methodologies," says Greaves.

Fitch operates Rap CD as an application services provision where it hosts the technology, collecting its subscribers' deal information from the managing banks, running the valuation and analysis, and delivering reports that include key tranche and reference entity sensitivities, correlations, CDS delta hedge equivalents and recovery rate sensitivities. Key advantages of the service are that subscribers do not have to input their deals or market data, nor do they need to have expensive in-house technology to get the reports, says Greaves.

With regulatory and market pressures growing for more frequent and accurate valuation of portfolios, the demand for price data and mark-to-market services, especially for more complex and illiquid products, is unlikely to let up. Newer entrants with innovative approaches are helping ensure that institutions have some options in where they get independent pricing.