NEW YORK (Reuters) - Ben Bernanke’s first news conference on Wednesday is a plunge into unknown territory for the Federal Reserve chairman and for computerized trading as well.
Computer trading programs face two dilemmas. There is no history of how security prices have reacted during a press conference with the U.S. central bank chief, and dialogue from the briefing will be spoken, rather than transmitted as text.
Computer-driven trading programs are designed to recognize text, so the nuances of Bernanke’s answers to reporters will be lost, or at least delayed, as humans intervene. That will make this inaugural conference a learning lesson for future Bernanke press briefings.
“It would be quite hard to get a huge amount of accuracy from a one-off, unstructured press conference,” said Rochester Cahan, a strategist at Deutsche Bank in New York who leads one of the major sell-side quantitative research teams.
“To trade that, algorithmically, would be quite hard,” said Cahan, referring to the software code that instructs computers what to buy and sell.
High-speed, computer-driven trading dominates Wall Street and has increased in other markets. But it’s still geared to pore over text in electronic form — corporate filings, the release of economic indicators, news stories or market data — in search of key words or patterns that might trigger trading strategies.
While the algorithms have become more complex and can easily find words or count them, such as the Fed’s use of “optimistic” and “upbeat” to describe the economy in its March 15 statement, parsing the shades of Fed speak can be tough.
For example, in March the Fed said data suggested the economic recovery is on a firmer footing. On January 26, it had said the recovery is continuing, although at a rate insufficient to significantly improve labor markets.
Algorithms may have 50, 100 or 200 variables that can instantly analyze how a market has performed in the past under similar conditions. They can decide in a snap what to do when a certain stock is rising, the industry that stock is part of also is rising and the overall market is rising.
Computers “will have gone back and looked at a million cases of where that’s happened and say what’s likely to happen next,” said Alfred Berkeley, chairman of Pipeline Financial Group Inc, an agency brokerage for buy-side clients.
The statistics might show that if those three things have occurred, the next trade is likely to be up, perhaps by 82 percent — a fact a human might recognize, but not as precisely.
“I might have an intuitive sense that the stock will go up, but I don’t have the precision of knowing 82 percent of the time, it’s going to go up,” Berkeley said.
“You can cut this data lots of different ways until you find these powerful correlations,” he said. “So you create these databases of circumstances.”
Algorithms are now so sophisticated they can interpret news stories in a way that a person would, focusing on positive or negative words to glean sentiment.
Although tricky, algorithms can even dissect a news story about a merger or an acquisition with adequate understanding, said Cahan, who is not a trader.
“That’s obviously quite hard for a computer to do, but it’s actually gotten to the stage they seem to be able to do that with some degree of accuracy,” he said.
Editing by Padraic Cassidy