High Frequency, Low Latency Technology briefing - Cape Town
On 15th March 2010 I gave a briefing in sunny Cape Town to a group of buy side technologists and business representatives, about the ways in which technology is driving trading activity. In particular in the High frequency (HFT) and low latency (LLT) arena. There was a great deal of interest in this highly technically oriented area, especially in trying to understand how the South African market can prepare for the impending arrival of the JSE's exchange platform in Johannesburg (it is currently in London, at the LSE) .
There is a distinction between HFT and LLT which is not often made but which is hugely important, especially in a market where the distance to the exchange makes low latency trading almost an impossibility. High Frequency Trading relates to the real time analysis of high frequency data and decision making on that data. This data, especially in short time windows exhibits a more statistically significant degree of regularity, and is more predictable that end of day, or general intra day data at low frequency. The importance then is in fast processing of data and identification of patterns rather than fast execution.
Low latency trading relies also on high frequency data, but furthermore works on a "first past the post" principle, since the trades that are made are often simple and can only deliver profit for a short period of time. In this case, the system with the fastest decision and execution time wins. this Ultra Low Latency trading requires Direct Market Access and the reduction of physical distance between the execution venue and the trading system to as small as possible (this mitigates the basic "speed of light" limitation). ULLDMA systems, as they are called, must be co-located with the exchange, and technology (networks, processing, memory) must be at the highest possible end of the spectrum. This leads to an inevitable arms race.
ULLDMA may not yet be available in the South African market, but the technology for HFT is. These systems typically require high end processing of data. The NYSE publishes over 1,000,000 messages per second, and an HFT system needs to be able to filter, clean and aggregate that in real time. To do this data needs to be in memory, immediately available and filtered as a stream not post hoc. Complex Event Processing technology is often the choice for such a task.
CEP technology has the capability to process an event stream as the stream passes through the engine, on the basis of real time events. Furthermore it can simultaneously process many such streams and look for correlated events accross them and can apply decision rules based upon them (i.e if IBM stock goes up 5% in a 10 second window against its 2 hr WMA, and Oracle goes up 3% in a 15 second window against its 1hr WMA, then buy Apple) . Sybaseand Oracle all have high end CEP offerings.
Such technology by itself can for the basis of a HFT system, but high end systems also include in memory caches for market and reference data, and high end parallel processing machines for analysis or real time pricing. Data cache technology (Celoxica) .
Michael Roberts, the head of South Africa for Excelian, kicked off the presentation with a review of the impact of regulation that would effect the buy side.
I followed up with an overview of the rationale behind the arms race, and the competing technologies that are under deployment by firms who are active in the HFT, ULLDMA space. The presentations are available Here.