The rates trading group at Mitsubishi UFJ Securities, part of Mitsubishi UFJ Financial Group (the largest financial institution in the world in total assets), is putting new technology behind its approach to relative value pricing and dissemination of market data. The group, which was formed in March 2006 and now employs 50 people, is installing a real-time tick database and low-latency price distribution technology."For years, we just used closing price information [for relative value pricing], but that's not anywhere near enough information about how the market's moving," says Greg Bufton, director of the rates trading group at Mitsubishi UFJ Securities. "When it comes to actually putting on positions, it's good to know yesterday's close price, but you also need to know where the price has moved over the last hour and the last five minutes."
The group will soon begin using Sybase's RAP - The Trading Edition time series database for the storing of tick data. "Our derivatives prices are going to be in-house, C++ functions; the latest thinking on how a discount curve should be built, that's all proprietary stuff," notes Bufton. "But we can store the information we get from that in the database, so that we can go back and replay our price at any moment in the past. We wanted to have a database that would be strong, robust and fast enough for us to be able to store large quantities of data we're creating from our front-end prices, to be able to play it back later," Bufton says.
Mitsubishi staff tested a number of database products, and found they were all fast. Sybase's RAP product was chosen because it required half the storage capacity of many of the others. (Sybase reaps storage efficiencies from its column-based format.) The Sybase product also came with a data filter and performance monitoring software from Aleri. "You can watch the database through Aleri, see where the bottlenecks are, and find problems that have yet to arise," Bufton says. "New databases can fall over completely; this is like having a flashing light. You see the light flash a long way away and prevent accidents."
Bufton also liked the Sybase product's combination of relational database and time-series database characteristics. "A relational database lets you do a search on the database universe and come back with a handful of records," he explains. "Relational databases are very good if you're looking for a particular entity. But if you want to grab a year's worth of tick data updates, it would take forever to produce it, to pull back every single one of those updates. Whereas a time series database stores everything and lets you grab a year's worth of data in one pass." Bufton also says there's a large pool of quant developers out there who are familiar with Sybase SQL.
Mitsubishi's rates trading group is also implementing data handling technology from Gissing Software to distribute price information from Reuters, Bloomberg and other providers in a low-latency fashion. "We use Gissing as a real time data network," Bufton says. "The Gissing stuff is fun because we've got a lot of really fast hardware on the trading floor and we've got apps that generate large quantities of data. We realized that you don't need to broadcast everything you calculate, you just need to broadcast the inputs for the calculations." For instance, if traders are pricing products off spreads, for example, they can simply broadcast any changes in the spreads. Such shortcuts make the sharing of data, especially among far-flung locations such as Tokyo and London, faster and more efficient.