CHAPTER 1
An Introduction to High-Speed and High-Frequency Trading*

I'm so fast that last night I turned off the switch in my hotel room and was in bed before the room was dark.

—Muhammad Ali

In early 2009, with markets fresh off of a harrying near-Depression experience, news reports began to circulate that among the few winners in financial markets in 2008 was a new breed of trading firms—so secretive as to make quant trading shops look like glass houses—called high-frequency traders (HFTs). It didn't take long for some in the press, political and regulatory circles, and even in the financial industry to begin telling a highly biased, basically fictional tale about high-frequency traders.

Following these stories (which immediately prompted a chorus of cries of “no fair”) came an unfortunate incident involving a programmer, Sergey Aleynikov. Aleynikov had left Goldman Sachs to join a then-newly launched HFT firm called Teza (which itself was formed by former Citadel traders). He was arrested in early July 2009 and accused of stealing code from Goldman to bring with him to Teza. What was most alarming to the public about this case had nothing to do with Aleynikov, Goldman, or Teza (intellectual property theft cases are almost never of interest to the broader public). The prosecuting attorney—in an effort to add weight to Goldman's allegations—said that the software that was allegedly stolen could be used to “manipulate markets in unfair ways.”1 This was eye-catching for many, ­because it linked high-frequency trading with market manipulation. Aleynikov ended up being convicted, but had his conviction overturned and vacated by an appeals court after almost three years of appeals and jail time. But the damage was done, and the stage was set.†

On May 6, 2010, U.S. equity markets collapsed and recovered dramatically. There was more than a 1,000 point drop in the Dow Jones, with about 600 points of the drop occurring in a five-minute period that afternoon. This was followed by a fierce rally, which wiped out most of the 600 point loss in only 20 minutes. The high-speed nature of this meltdown and recovery came to be known as “the Flash Crash.” The Flash Crash was widely blamed on HFTs, though often for contradictory reasons. Some claimed that HFTs caused the crash by virtue of their trades. Others claimed that HFTs caused the crash because they stopped trading once the markets became too panicked. We will address these claims in more detail in Chapter 4, but for now, it suffices to say that the Flash Crash was a major contributor to negative popular opinion about a topic that almost no one understands.

According to the Aite Group, HFTs account for a little more than half of global equity volumes, about the same percentage of futures volumes, and about 40 percent of currency volumes. In equities specifically, Aite estimates that HFT's share of trading is highest in the United States (again, a little over half), more than 40 percent in Europe, and almost 20 percent in Asia.2 While there are various estimates of the exact amount of trading that comes from HFTs, no further evidence is needed to demonstrate that this kind of trading is a critically important topic to understand for any electronic market.

So what is high-frequency trading? It turns out that HFT is not a well-defined, ­homogenous activity. There are multiple kinds of high-frequency traders. But a definition that probably contains most of these kinds of traders is as follows: High-frequency traders (a) require a high-speed trading infrastructure, (b) have investment time horizons less than one day, and (c) generally try to end the day with no positions whatsoever. The fastest HFTs (sometimes referred to as ultra-high-frequency traders, or UHFTs) will no doubt scoff at the notion that someone who holds positions for as much as six and a half hours should be considered high frequency. But there is an important distinction between overnight risk and intraday risk, as most news comes out when markets are closed. Any further attempt to narrow down the holding period of an HFT strategy would be arbitrary: What makes a holding period of one second “HFT,” while one minute is not? Furthermore, our definition specifies that the strategy should require a high-speed infrastructure.

It is worth knowing, however, that HFTs share the high-speed trading infrastructure mentioned above (and described in detail in the next chapter) with many kinds of algorithmic traders. And high-speed infrastructure does not have only one speed. As we will see in the next chapter, the challenges facing engineers of such infrastructure are substantial, and in few instances does any industry standard exist to meet those challenges.

Within the algorithmic trading community, people tend to think of the users of high-speed infrastructure as falling into four categories: UHFTs, HFTs, medium-frequency traders (MFTs), and algorithmic execution engines. But all of these approaches tend to share commonalities in terms of the definition above (though algorithmic execution engines generally attempt to help acquire longer-term positions, the algorithm itself is usually not interested in what happens tomorrow). As was famously said by Supreme Court Justice Potter Stewart in 1964, “I shall not today attempt further to define the kinds of material I understand to be embraced within that shorthand description; and perhaps I could never succeed in intelligibly doing so. But I know it when I see it. . . .” He was talking about hard-core pornography, of course, but the exact same sentiment applies (in more ways than it should) to HFT.

This act of definition is important for a variety of reasons. First, it allows us to have a common footing when discussing the topic. Second, there are implications to any definition, including this one. This definition of HFT implies that there are several important characteristics of HFTs. First, since HFTs tend to end the day with no positions (flat, in industry lingo), their buying and selling activity tends to exactly offset. However many shares or contracts were bought of a given instrument had to have been sold as well; otherwise, there would be a net position at the end of the day. Second, since there is a desire to be flat at the end of the day, an HFT strategy likely does not seek to accumulate large positions intraday.

Accumulating large positions incurs large market impact costs. We define market impact as the amount that your own trading moves the price of the security in question. Typically, the larger the size of a desired trade, the larger the market impact. Unwinding such positions would cause further market impact costs. With an intraday holding period, there is not enough price movement on a typical day to offset the market impact of buying, plus the market impact of selling. Furthermore, since HFTs cannot accumulate large positions, and since price movements during the trading day are of limited magnitude, HFT strategies generally have fairly low profit margins. They must pay the same kinds of costs as other investors and traders—commissions, market impact, and regulatory fees, for example—but they participate in relatively small price moves. There are economic incentives associated with higher volumes of trading (such as exchange rebates for providing liquidity or cheaper commission rates from brokers), but the fact is that margins remain very low, and the aggregate cost of technology, commissions, and regulatory fees is usually a large multiple of the net profits after these costs.

Multiple sources have pegged the profitability of an HFT strategy in U.S. equities (in relatively good times) at approximately $0.001 per share (one-tenth of a penny). It is instructive to compare this to SEC regulatory fees of $22.40 per million dollars sold. This fee translates to about $0.006 per share for a $70 per share stock, though it applies only to sales. Since HFTs tend to buy as much as they sell, we can divide the fee by two to get a figure that applies to every HFT transaction, and we arrive at a typical SEC fee of approximately $0.003 per share, which is about triple the profit margin for a typical U.S. equity HFT.

It is also worth understanding that an HFT who trades 100,000,000 shares per day is responsible for more than 1 percent of U.S. equity volumes. His profit, at $0.001 per share, is about $100,000 per day, which comes to $26 million per year. If we multiply this by the 50 percent estimate of the share of HFT volumes versus overall market volumes, we get $1.3 billion in total trading profits for the entire HFT industry for an entire year in the largest equity market in the world. These are lofty numbers, but they are actually just the revenue side of the ledger.

The reality is that it also costs millions per year for every firm that tries to achieve these revenues, and there are a large number of firms that fail. And since there are a large number of firms competing, even a successful one that accounts for, say, 5 percent of the U.S. equity market's volumes (which would make it an extraordinarily successful outlier) is making something like $65 million per year in revenues, before accounting for the extensive technological, compliance. and human resource costs that are required to compete. It is worth noting that the TABB group's latest estimates of HFT profits in U.S. equities for 2013 show similar results as our back-of-the-envelope calculation: that the entire industry's profits in...