Drilling Down to the Cash Flows

Sumitomo's risk system builds cash flows from thousands of instruments into a global river of data.

In September 1995, Sumitomo Bank was facing what seemed at first like an insurmountable problem—how to implement an enterprise-wide risk management system.

Throughout the 1990s, the landscape for complex derivatives transactions in the global financial markets expanded at a rapid pace. Sumitomo Bank was no exception, since it was also increasing its derivatives activities within its international trading group and with customers. The Bank for International Settlements announced it was requiring global banks to calculate capital for market risks by March 1997. Although Sumitomo had sound risk management systems in place at the time, the risks were not calculated consistently across the enterprise, and consolidating the risks on a global basis was a daunting task.

A major risk management system capable of managing the bank's vast operations would have to be built, more than a dozen legacy systems would have to be connected, and the data would have to be accurate and free-flowing at every turn—an unprecedented feat in 1995.

According to Takashi Maruyama, senior vice president of market risk management, Sumitomo began the process by taking a careful inventory of the approaches available. One possibility was simply to add risk management functionality onto Sumitomo's existing trading systems. This was unattractive for several reasons. Trading systems have so many native functions that it's quite difficult to add robust new risk management capabilities without compromising performance, and for each new product, the entire risk management system would have to be rebuilt on top of the existing trading systems.

Another possibility was to develop a system that would use the output from all of the bank's legacy systems to calculate risk figures. Such an approach certainly would have been the most convenient and the cheapest way to get some kind of summary risk number—but it, too, was fraught with problems. Automating the process would be impossible, since each legacy system was unique. And manually aggregating risk figures from all of the legacy systems would introduce an unacceptably large margin of error. Moreover, it would be difficult or impossible to do stress-testing on each of those systems. Once the results had been summarized into aggregate risk statistics, the information would remain static. It would also be difficult to obtain accurate results by performing stress tests on the aggregate summary statistics—particularly for positions such as options, with significant first- and second-order risk. And it would be impossible to track back to the applicable source systems to determine the positions and cash flows responsible for any significant changes in risk levels. Additionally, it would have been virtually impossible to calculate the impact of market risk on credit risk and liquidity risk.

Mixed fruit

Maruyama says Sumitomo decided that the only way to resolve the problem was to develop a database capable of capturing the cash flows for every transaction on Sumitomo's books. For that to work, legacy systems would first have to generate the cash flows in a consistent way. The traditional approach would have been simply to map fields from each legacy system. But cash flows from, say, a foreign exchange options system and cash flows from a swap system would not be stored in the same fields and in the same manner. Consequently, they wouldn't have the same common denominators. Apples would be mixed with oranges—and pears and bananas. "Typically, most enterprise-wide risk management systems set up a database for each type of security and asset class individually and then aggregate the risk for the different instruments,” says Maruyama. "But unless you understood how those systems were aggregated, you wouldn't have a common denominator, because each cash flow would have been estimated in a different way and stored in a different database structure.”

The goal, then, was to develop a system that looked only at the individual cash flows for all transactions, regardless of the instrument type, and combined various streams of assets into cash-flow rivers that could be easily aggregated. Without such a system, it would still be possible to determine the cash flows from, say, loans—but not the cash flows for upcoming interest payments or the cash flows from money lending or swaps. With the new system, there would be one vanilla set of cash flows.

Another goal was to deal with exotic types of transactions. New exotic transactions were generated every day. Some types might survive, some might not. Yet the risk manager has to incorporate these cash flows into a comprehensive risk management system, in a timely manner without significant additional cost.

At the back of Maruyama's mind was another goal. He knew the bank would ultimately have to consolidate its credit and market risk reporting, as well as its liquidity risk. The database designed in 1995 would have to provide the flexibility necessary to allow this to be done sometime in the future. It would be inefficient to build two different databases. Therefore, the market risk management database and the credit risk management database would have to be unified.

"The sort of issues Sumitomo had been struggling with had already been resolved at BT. This enabled Sumitomo to reduce the risk of project failure and the necessary implementation time.” — Ed Berko

After approaching a number of vendors, Sumitomo decided to go with a subsidiary of one of its U.S. rivals, Bankers Trust. The private trading house had spun off some of its own systems into a subsidiary called IQ Financial Systems Inc. One of IQ Financial's main products, the RISK IQ enterprise-wide risk management system, had been developed by Bankers Trust, based on its proprietary risk management architecture. "The sort of issues Sumitomo had been struggling with had already been resolved at Bankers Trust,” says Edward Berko, president of IQ Financial. "This enabled Sumitomo to base its risk management system on the successfully demonstrated risk architecture developed by Bankers Trust, and, as a result, to significantly reduce the risk of project failure as well as the necessary implementation time.”

During its initial vendor search, a number of software companies said they were busy developing the kind of consolidated database Sumitomo was looking for. But only IQ Financial presented an architecture that had been proven at Bankers Trust. Sumitomo signed on with RISK IQ, and spent the next 12 months modifying and implementing the system to go live before the March 1997 BIS deadline. "Sumitomo was unbelievably dedicated to this project,” remembers Catherine Osti, RISK IQ product manager at IQ Financial. "Sumitomo staff worked with us on a regular basis to develop the required specifications. Sumitomo Bank was truly our development partner in extending RISK IQ's functionality.”

The biggest job was developing the links between the company's two-dozen legacy systems and RISK IQ. The key to the process was getting the data out of the legacy applications in the most efficient manner possible, to populate the consolidated risk management database. The traditional approach was to make alterations in the existing legacy systems, so that the data could be output to the risk database. But Sumitomo decided that this involved more problems than it was worth. Its goal was to set up a "non-invasive” way of extracting the data.

IQ Financial's solution was something called a transfer manager, which analyzed data coming out of the front-end legacy system. The transfer manager would sit on top of each legacy system and map transaction data from those systems to predefined fields in RISK IQ, which would then generate the cash flows for each product in a consistent manner. "The transfer manager is the heart and soul of the system,” says Maruyama. "It allows us to consolidate the risks across the firm into one common denominator.” The first phase of implementation required transfer managers to be created for Sumitomo's trading, foreign exchange, general-ledger and derivatives legacy systems.

There were many other adaptations that needed to be completed as well—security functions, batch-processing enhancements and so on. Sumitomo, for example, needed a system that permitted data to be consolidated from several different global offices. The global transfer manager extracted data from a local site and brought it into the global database. Sumitomo then recalculated global VAR based on cash flows from the global database using consistent valuation models and Sumitomo's official market rates, to ensure a consistent enterprise-wide evaluation of risk. This approach turned out to be far superior to adding up VARs from each location or each region.

Work in progress

After the global system went live in March 1997, the team started planning for a second phase of development. The most urgent problem was developing a system feature to enable users to correct bad data inputs. Inevitably, a few of the thousands of inputs made to the consolidated risk database every day would have to be corrected. This was a maintenance nightmare, requiring the systems team to correct and restore databases, and then re-run the analytics to get a revised overall VAR number. Ironically, this would have been relatively easy to do if Sumitomo were developing manual risk summaries in a more primitive system. But in Sumitomo's more ambitious, complicated system, a new approach was necessary.

The ultimate solution was to develop 10 new databases—one for each of the 10 previous business days—allowing Sumitomo to back-test several days' worth of data. If there was a mistake, risk managers could make adjustments in the historical data with full audit capabilities, so that all changes would be documented. "Since the VAR calculation is based on accurate transaction and market data, incorrect data can skew the VAR and back-test results, providing an incorrect assessment of the risk and accuracy of the VAR model. Therefore, it is very important to be able to edit the historical databases with full audit controls and recalculate the VAR and back-test results,” says Berko.

Gregory Lucas, vice president for risk management at Sumitomo Bank, regularly reaps the benefit of the RISK IQ. Now that the system is up and running, one of his most important jobs is monitoring the accuracy of the raw market and transaction data in the RISK IQ enterprise system. He regularly reviews the data that has been loaded into RISK IQ and compares it with the source data from the legacy systems to assure that there are no discrepancies. "I query RISK IQ for all the data between two specific dates and then use a RISK IQ tool to compare this data with the source data in the legacy systems,” he says. "And RISK IQ provides me with the flexibility I need to dynamically correct any data discrepancies and to automatically re-run all of the necessary risk statistics and back-tests on-line. One of the strengths of RISK IQ is its relational database. The painstaking process of scrubbing data can be automated using off-the-shelf database and spreadsheet programs. Furthermore, the database permits one to dissect VAR into individual components in order to identify the primary risk contributors and analyze irregularities.”

An enterprise system

In addition to calculating the bank's global VAR, the system provides a means for the bank to manage liquidity risk on an enterprise basis. Utilizing the relational database, risk managers at each location can summarize the liquidity risk across all products, based on the transaction cash flows from the database. These figures are reported to members of the head office, who reconcile the report to the global database to ensure consistency throughout the system.

Another challenge met by Sumitomo Bank's risk system involved enhancing performance to achieve global consolidation of data and generation of VAR within a 24-hour period. This enables Sumitomo to make any necessary portfolio adjustments to mitigate unusual or unexpected changes in risk based on the results from the prior day. This is an amazing achievement when one considers that Sumitomo has several million cash flows at the transaction level in RISK IQ's consolidated global risk database. And it permits Sumitomo to zero in on the exact causes of any unusual changes in the daily risk—including the individual transactions.

Going forward, Sumitomo is considering a new set of improvements as well. Although the bank has focused primarily on market and liquidity risk, IQ Financial is further enhancing the integrated market and credit risk management capability in RISK IQ.

"The key strength of RISK IQ is its plumbing and the consistent data across all products and locations,” concludes Lucas. "Risk managers can employ the analytical tools they feel most comfortable with, including valuation models such as historical or Monte Carlo simulation, and user-definable stress tests. And the risk results are obtained within such a short period of time that they can be used to make the necessary decisions and portfolio alterations.”