HFT - High frequency trading accounts for 80% of all trades..
By some estimates, 90 percent of quotes on the major exchanges are canceled before execution. Many of them were never meant to be executed; they are there to test the market, to confuse or subvert competing algorithms, or to slow trading in a stock by clogging the system—a practice known as quote stuffing. It may even be a different stock, but one whose trades are handled on the same server. On the Internet, this is called a denial-of-service attack, and it’s a crime. Among quants, it’s considered at most bad manners.
[snorf]
overview from article..halfway down the page..
High-frequency traders make money in a vacuum, grabbing for pennies that appear and disappear like the virtual particles of quantum field theory. Their goal is to end each trading day “flatâ€â€”out of the market, their profits safely in the bank. Depending on their model, they can do well winning as little as 55 percent of their trades. They are continuously testing prices, looking for patterns and trends or the chance to buy something in one place for $1 and sell it somewhere else for $1.01, or $1.001. Sometimes they aren’t even looking to make money on the trade itself. Under the “maker-taker†model, some exchanges offer tiny incentive payments, or rebates, for posting a quote (to buy or sell a stock) that results in a trade. The exchange charges the other side in the trade, the taker, a slightly higher fee and collects the difference. So an algo can buy a stock, earn a rebate, then sell the stock and earn a rebate for that too.
All of this is governed by algorithms whose lifespans can be as short as a few weeks. Sometimes an algorithm does something as simple as look for a stock that ticks up in price several trades in a row. A “momentum†algo would buy the stock, expecting the rise to continue. A “mean-reversion†algo would sell, expecting a drop back to average price. They might both even be deployed by the same firm. Over the course of a minute, they might both be right.
One common algo strategy is to look for pairs of stocks whose prices are historically correlated. The canonical examples are the stock prices of oil companies, which rise with the price of crude, and those of airlines, which do the opposite. But they may not move all at the same time, so one strategy is to buy or sell the one that’s trailing and wait for it to catch up. Similarly, “derivative†equities such as options and futures may get out of equilibrium with the underlying stocks. Some algorithms are “market makers†in a stock—they attempt to buy at a low bid price and quickly sell at a slightly higher asking price, pocketing the difference, or spread. The people who did this used to be called specialists, and it was a nice living when spreads were an eighth of a dollar. Since the New York Stock Exchange instituted “decimalization†in 2001, spreads have gone down to a penny or two, meaning you have to trade a lot more stock, a lot faster, to make the same amount of money. It’s no place for a human being.
http://www.wired.com/business/2012/08/ff_wallstreet_trading/all/
| 