High-Frequency Trading

"High frequency trading" is perhaps one of the most-abused terms in the market today. (Everyone wants to be doing it, but not many people can give a good explanation of what it is.) It can be regarded as a sub-category of algorithmic trading characterized by two things: Here's a table with one possible categorization of algorithmic trading systems, based on the average trade duration.

Category Reaction Speed Average Trade Duration
Very Low-Frequency up to several hours 1 week or more
Low-Frequency up to a few minutes 1 day to 1 week
Medium-Frequency up to a few seconds 10 minutes to 1 day
High-Frequency 100 milliseconds or less 1 second to 10 minutes
Very High-Frequency 1 millisecond or less 1 second or less

This categorization is based on average trade duration.  There is no need to explicitly categorize based on reaction speed, because average trade-duration typically governs the requirement for reaction speed.  As a general rule, the shorter the duration of the trade, the more accurate the timing has to be in order to overcome transaction costs.  So, for example, it is not possible (unless you want to lose lots of money) to have an algorithm with an average trade duration of 1 second that reacts to market data with a latency of 1 minute.   (Note that this is just one possible categorization - to my knowledge there is no official version of this.)

Costs required for infrastructure increase dramatically as you go down the rows in this table.   An individual without prior experience trying to get started as an algorithmic trader typically cannot go beyond the medium-frequency category in this table.  High-frequency or very high-frequency algorithms require significant investment in infrastructure (co-located servers, expensive trading software, and at least one or two good programmers who can handle high-speed real-time systems, not to mention the time required to get all of this set up).

 Next: Typical Infrastructure            Back to Index