Man Vs. Machine: Pros and Cons of High-Speed Trading
By: There are algorithms for all different kinds of trading strategies
High-frequency trading is the main reason volume on the NYSE has increased 150 percent in five years
More than half of all US stock trades made each day are the result of high-frequency trading.
Regulators are trying to figure out whether this form of high-speed buying and selling contributed to the May 6 "flash crash," in which the Dow Jones Industrial Average dropped 1,000 points in less than an hour, before recovering.
What Is It?
The two essential elements of high-frequency trading are an algorithm that can accurately identify a pricing mismatch or trading opportunity, and trading systems that are lightening fast—trading speeds are measured in milliseconds (thousands of a second), and, increasingly, in microseconds (millionths of a second).
Beside speed, the other characteristics of high-frequency trading are:
- High Volume: Traders turn over positions thousands of times a day, hopefully making a small profit on each trade
- Short Term: High-frequency shops do not usually hold positions overnight
- Usually Done on a Proprietary Basis: Participants use their own money.
Not all high-frequency shops are the same. Some are "passive market makers." They provide bids and offers to the markets and make money by playing the spread and collecting a rebate.
Here's an example: A high-frequency trader (HFT) who has a layered series of bids and offers on IBM might buy 100 shares of IBM for, say $128.59, and sell it a fraction of a second later for $128.60. The profit: $1.00 (100 shares at a profit of $0.01).
But there's more. Exchanges pay rebates to provide liquidity (offers to buy and sell stock). An exchange might pay, say, 14 cents per hundred shares.
So in this example, the HFT would make $1.00 trading the spread, plus a 14 cent rebate for providing a bid (offer to buy) and another 14 cents for an "ask" (offer to sell).
Total profit: $1.28. Do this millions of times a day, and the pennies add up.
Other HFTs try to add alpha (outperformance) by using specific trading strategies.
There are algorithms for all different kinds of trading strategies. The most common is statistical arbitrage, where traders are short one security and long another, based on historical performance. One example is "pairs trading:" stocks in similar industries (Wal-Mart and Target , for instance) tend to move in tandem; when the correlation breaks down, pairs traders sell the outperforming stock and buy the underperforming stock, believing that the correlation will revert to the mean.
Exchange-traded funds (ETFs) can also figure prominently in arbitrage. For example, one strategy might seek to identify a price discrepancy between an ETF and the underlying basket of stocks. When the ETF price is higher than the value of the basket, the HFT sells the ETF and buys the underlying stocks, and vice-versa.
These pricing inefficiencies may only last for a fraction of a second! "When you have opportunities as small as a tenth of a cent per share, they tend to come and go very quickly," says Manjo Narang, founder and CEO of Tradeworx, a high-frequency trading firm in Red Bank, New Jersey.
There are also algorithms for trading volatility, where traders make bets on relative price movements of securities (large or small), and for trend following (sometimes called "sentiment trading" or "event-based modeling") where trader reaction to economic and market news is analyzed and broken into algorithms that can execute trades at the moment news comes out.
How important are algorithms? They are everything—so important that it was a major story on trading desks when a Goldman Sachs trader, Sergey Aleynikov, was arrested by the FBI on allegations that he stole Goldman's proprietary trading code—a mere 32 MG of data!
How Prevalent Is It?
Total daily volume in all stocks listed at the New York Stock Exchange went from about 2 billion shares a day five years ago, to an average of about 5 billion shares a day today. That's a 150 percent increase, almost all of this gain is due to HFT strategies. High-frequency trading now accounts for about 56 percent of trading volume, according to Tabb Group, but Tabb notes that this figure includes market makers. Five years ago, it was practically nothing.
Who's trading? Here's the latest breakdown of daily volume (source: Tabb Group):
- High-Frequency Trading: 56 percent (includes proprietary trading shops, market makers, and high-frequency trading hedge funds)
- Institutional: 17 percent (mutual funds, pensions, asset managers)
- Hedge Funds: 15 percent
- Retail: 11 percent
- Other: 1 percent (non-proprietary banking)
How profitable is high frequency trading? There's no doubt this can be profitable, but profit margins are tight.
The Tabb Group estimates that there are roughly 400 HFTs. Of that, there are roughly 150 HFTs that trade U.S. equities. They estimate that the gross trading profit for all HFTs in 2010 will total $5.6 billion (not including brokerage and SEC fees).
That works out to about 0.2 cents per share gross profit, but once brokerage and SEC fees are deducted, net profit was about half that.
Tradeworxcame to somewhat similar conclusions. In a filing with the SEC, they estimated net profits of HFTs at $2 billion a year, once brokerage and SEC fees are deducted. Their estimate of net profit margin is 0.1 cents per share.
With profits like that, you need to trade big numbers. It means, on average, you need to trade 10 million shares a day to make $10,000. Lower volume means less opportunity to make money.
Sound like a lot? It's not peanuts, but compare that to the billions that mutual funds charge.
Is Something Wrong?
Does high-frequency trading give investors an unfair advantage because they are supposedly smarter and faster? The trading community has always sought advantages—whether it was whispered information, or having someone posted on the floor of the exchange, or just being closer to a telephone or telegraph a hundred years ago, or having access to a Bloomberg terminal a decade ago.
"On average, they (HFTs) have reduced our transaction costs and saved the investors in our funds a lot of money," says Gus Sauter, Chief Investment Officer of Vanguard, one of the country's largest mutual funds.
But critics believe there are real issues. Among the complaints often heard against HFTs:
- It adds no real "economic value." "They're just scalping pennies! They're not trading based on any fundamental economic reason for owning or selling a stock!"
This is true, but not interesting. People have bought and sold stocks for short-term advantage, with no interest in long-term fundamentals, since the dawn of trading. HFT just moves the time frame up to fractions of a second.
- Certain trading strategies are a form of market manipulation or may otherwise harm long-term investors.
Maybe, but we have very little information on who uses what types of trading strategies.
While most HFTs seem to employ old-fashioned arbitraging strategies, the SEC has voiced particular concerns over two other types of trading strategies.
The first is order anticipation strategies, which use pattern recognition software to determine where large buyers or sellers are, or the use of orders to "ping" different market centers. The goal: figure out where stocks are going, then buy low and sell high, or short stocks that look like they are going down.
Here's an example: Suppose you are a mutual fund looking to buy 100,000 shares of IBM. Suppose IBM was trading at $125. On an exchange, an HFT will likely have layered bids and offers in at IBM several cents from the last price. In this case, let's assume there are bids and offers in from $124.95 to $125.05.
You (the mutual fund) have a limit order to buy 100 shares at $124.99. It gets taken, the HFT will immediately put in another offer to see if it gets hits again, and once you get three or four hits, the HFT can make an assumption that there is a fairly large buyer in the market. Then the HFT will see if they can buy at $124.98 from another source and try to sell it to you for $124.99. This strategy then repeats.
