Citigroup revamps trading algorithms amid 'arms race' in stocks

Citigroup Inc. is overhauling its platform of trading strategies in an attempt to grow its share of U.S. equities from 14%, according to Young Kang, global head of algorithmic products at the firm.

The New York-based broker is introducing a new set of algorithmic strategies, as well as technology focused on speed and geared to an electronic marketplace with dozens of venues. Citigroup's almost 1,000 clients will make the switch to the new trading products by May 1, when the old system is retired.

The initiative, run by Mr. Kang, is designed to woo business away from rivals including Credit Suisse Group AG and Goldman Sachs Group Inc. The global platform, called Citi Algorithmic Trading Strategies, has been used by the firm's own traders for 11 months and is now available to clients in North America, Europe, Asia and the Middle East.

“Algorithmic trading is a complete arms race, and brokers must keep improving their offerings,” said Sang Lee, founder of research firm Aite Group LLC in Boston. “Citigroup has a lot of clients that are traditional asset managers driving a lot of volume” and may now be after more trading from a broader range of institutions across more geographic areas, he added.

Steve Swanson, head of global electronic trading at Citigroup, said he expects the new platform to help lift the broker's volume toward 20% of the market from its current 14%. He said 20% is Citigroup's goal across all of its equities businesses.

“This has been our biggest project for the last year and a half,” said Mr. Kang, who joined Citigroup in April 2007 from hedge fund SAC Capital Advisors LP, in Stamford, Conn., where he handled algorithmic strategies and trading technology and helped create Citigroup's most-used Dagger algorithm. Dagger represents about a third of Citigroup's algorithmic trading volume. “The new platform for algos will deliver a higher quality of execution to clients,” he said.

A global group of 20 quantitative analysts and more than 50 developers, overseen by Mr. Kang, built the algorithms and the technology supporting the new platform, which has been used on Citigroup's cash equities and program trading desks. The strategies have been tested in recent months by some hedge funds and institutional clients.

The system analyzes the trading history of a given stock, current market conditions and how venues execute orders when making decisions, Mr. Kang said. These changes have improved the performance of Dagger, which actively seeks liquidity, and the firm's other algorithms including what he calls the “utility” strategy that seeks to achieve the volume-weighted average price for institutional orders. VWAP is a stock's average price over a certain period, weighted for the number of shares executed at different price levels.

“Some people think VWAP is a utility and is ubiquitous, but even VWAP can be improved by doing things right,” Mr. Kang said. Citigroup was able to execute orders 4% closer on average to the market VWAP using the new algorithm and the “variability of performance” fell 41%, with fewer transactions occurring further away from the average, he said.

Some of the changes to Citigroup's algorithms were made in response to the dominance of high-frequency trading, or computer-based executions by firms that trade a lot of volume at speeds usually less than a thousandth of a second. This trading, which encompasses multiple strategies, accounts for 61% of equities volume, up from 35% in 2007, according to research firm Tabb Group LLC in New York. Mr. Swanson said he believes high-frequency trading represents less than 40% of the market.

“Because of how the market trades today, with fragmentation and gaming, you need to protect the order more than before and grab liquidity faster,” Mr. Kang said. “You don't want to signal you're aggressive or interested in a particular name.”

It's increasingly important to know “how each venue works” and to design tactics to try to maximize liquidity while minimizing the “chance of an order being picked off or gamed” by firms trying to take advantage of institutional orders, Mr. Kang said. He added that strategies must figure out “when to trade and, perhaps more importantly, when not to trade, along with where to seek liquidity and where not to.”

Citigroup is counting on attracting more orders from large investment firms because their traders will have “better experiences” using the firm's algorithmic strategies and be able to “verify their positive experience” by comparing the executions they get relative to particular benchmarks to those from other brokers, he said.

As part of the effort to improve executions and avoid releasing signals about large transactions into the market, Citigroup's algorithms have been optimized to seek out more trades in Citi Match, its dark pool, or off-exchange venue, which attracts a lot of orders from individual investors, Mr. Swanson said. Citi Match, unlike some competitors, does not allow high-frequency traders into its pool, a restriction that benefits clients, he added.

Citigroup is also making changes to ColorBook, the smart router from Lava Trading, a unit within the broker, to execute orders faster. Smart routers are used by brokers to craft an execution strategy for the smaller orders generated once an algorithm or trading desk has decided how to transact a stock.

ColorBook currently executes orders in 700 microseconds, or 0.0007 second, Mr. Swanson said. The firm is trying to reduce that to 200 microseconds.

“We are racing to make all our systems as fast and capable as possible,” Mr. Swanson said. “Speed is important because it determines whether or not you execute,” he added. “You now have to touch flow as fast as possible.”