High Frequency Trading

There’s a lot of FUD about high-frequency trading. After reviewing much documentation on it, and discussing the math with a PhD in Applied Mathematics from MIT who’s been working in cryptography and numerical analysis probably for longer than you’ve been alive, I have begun to understand how and why it works for those involved, and how it places an unfair advantage in the laps of those who participate.

First, the computer hardware requirements are really special. The exchanges charge huge fees for participation, as that hardware needs to be placed at or near to the Exchange’s servers such that the hardware may be connected directly to the Exchange’s GbE (Gigabit Ethernet) network. The servers running a company’s HFT algorithms must be physically placed on the GbE links provided at the exchanges because retransmission of packets through physical layer links takes time. Latency in information on a trade of milliseconds will cause the probability algorithms running on the servers to become unstable, causing significant errors in the calculations of the buy/sell orders, their values, the bid/ask information, and other pertinent information. So Cisco and others have developed special low-latency switches that permit scalability on this exchange-provided GbE network.

The second piece of latency is in the operating system running on the servers. In our opinion, the HFT servers do not run a standard operating system, rather they would be running either a proprietary OS or pieces of an available open source or purchasable source OS, perhaps like Linux, where the OS could be modified to eliminate latencies and packet forwarding delays. (packets forwarded to other layers of software in the servers). The servers would all be running in parallel, because running a standard network topology where packets are forwarded would be far too slow.

You and I can’t afford these switches, servers, and specially modified operating systems, and we can’t afford the fees charged by the exchanges to be co-located on the Exchanges’ GbE links. Therefore, the playing field is not level. It is even not level for firms that do not have the IT smarts to develop the low-latency scalable networks and have predictive calculators running which have the answers done before the data arrives.

Whether the playing field should be level should be the subject of another post.

Having answers before data arrives is .key in HFT. You’ve heard about multi-processor chips from Intel, well HFT servers are exactly that, multi-multi processors which run algorithms on data that hasn’t arrived, so as soon as the data does arrive, the server with the correct answer simply puts out it’s precalculated answer, and the buy/sell order that has been precalculated is put in place. The other servers are off calculating the next one before the data arrives from the trade just placed.

By doing this, and placing orders in quadrants around the previous orders, one can collect a kind of topographical map of orders placed/filled or placed/not filled, and by doing so, one can predict the probable volumes of shares that are available at various price points. The map is kind of like a sheet that you’ve shook out over your bed, where there are billows and valleys, where some parts of the sheet will hit the bed before others. Once you know the characteristics of the air under the sheet, you can predict which parts will hit the bed first, and more importantly, next.

Now here’s the next one that’s unfair to us traders. If there’s only one HFT trader on any one given stock, that HFT is guaranteed to make money. So there are algorithms that evaluate how fast the sheet billows on any given stock, and these algorithms can determine how many HFTs there are participating. If there are more than your algorithm wants, your algorithm simply moves off to another stock. Also, the algorithms develop signatures, and if one of your algorithms evaluates the sizes, buys sells, and other characteristics, you can tell which company you’re competing with. In fact, you can figure out who you must not compete with in order to make money. That’s why almost all, and I will hazard a guess that all, HFTs make money. They know the competition.

There are numerous advantages given to HFTs on any given exchange, set up by Exchange rules. Just how this is accomplished is not for public view, but there are reports of several things given to HFTs to theoretically assist the Exchange in doing it's business. For example, the human market makers are gone, so computers have to have supplies of shares of stock to complete transactions. If an HFT provides these shares, the Exchange is grateful and may give that HFT a kickback. However, providing these shares gives the HFT an advanced view of the orders as well, so they may be able to place orders to make more than just the kickback.

At this time, there are no rules nor regulators nor regulations by which HFTs are controlled. It is up to the Exchange to police their own environment. How well this is done is unknown.