Conference Take-Aways: InVantage Group’s HIFREQ TRADE

Bob Giffords

Independent Banking & Technology Analyst

Is high frequency trading (HFT) just low frequency trading on steroids or something rather different? That was the question posed at InVantage Group's major new conference in London last week. I for one came away with the view that we're seeing a qualitative change in the market fabric.

With financial trading data rates growing at 100% per year, much faster than the global Internet but still trailing behind mobile data growth, Europe's first HFT conference was packed. Although the VIX volatility index fell from its peak of nearly 90 during the 2008 banking crisis to under 24 in December last year, financial trading volumes just powered on regardless. HFT is here to stay.

There were many memorable moments, like the sell side firm that is 100% automated in Europe. They employ no 'traders' as such, almost all of their interaction with the market is now managed by machines. There was the hedge fund that insisted algo latency was more important than network response times. At 200 microseconds for the algo, he couldn't afford the luxury of sponsored access and opted instead for direct exchange memberships. Another broker argued that smarter is more important than faster and understanding why things happen was key. Indeed there was a lot of discussion about explicability and ways to improve it in this spooky low latency world.

There was also a refreshingly constructive dialogue between European regulators and market participants over risks and opportunities. A consensus was emerging that liquidity pools need to ensure 'fair and orderly markets' even at these lightning speeds to protect participants. Regulators too need to join the technology arms race, if they are not to be left behind. Clearing agents called for minimal standards on pre-trade risk checking and flow control for DMA or sponsored access since there were gaps they could not address individually. Participants also highlighted the inherent conflicts of interest of some toxic flow. So yes, some areas were ripe for regulation, but there seemed overwhelming support for HFT because it had both democratised fragmented markets, making them more open, and tightly integrated them as well.

Storm Warning

One of the most interesting talks profiled a series of high frequency trading styles, illustrating the micro-bursts and volatility of orders, the high levels of open exposures and relatively low fill rates. One strategy set up around a couple of thousand open orders at start of day with a one-minute burst of tens of thousands of orders, and then kept revising them in smaller bursts throughout the day. Another strategy saw a swarm of orders and cancels at the end-of-day auction, a third had huge spikes of activity, tens of thousands of instructions in a mere second, with periods of relative calm in between. The average life of passive orders also varied from millisecond flickers, to exposures lasting hours, although price, size and even direction of trade might be constantly adjusted. Another speaker emphasized holding times for stocks, which might last seconds, minutes or hours and the spread of trades across trading venues that were highly sensitive to the quality of liquidity and hit rates. Holding times, he noted, affect market or factor neutrality and thus the amount of capital needed.

These were highly choreographed, but wonderfully varied basket trades. As markets lurched and fills rolled in, thousands of exposures would need to be rebalanced. I immediately started imagining how these might aggregate at the market level, and here the metaphor was one of weather: storms of orders sweeping across countries, sectors and asset classes, with micro-climates around individual instruments or related clusters.

The need for huge firepower to keep pace with the regime became immediately clear given the scale of exposures. Here high-core-count servers and GPUs are not a nice-to-have, but essential to keep pace with the microbursts. [See a review of Accelereyes GPU application in "Matlab's Racing Jacket" from the Q1-2009 issue of Automated Trader - Ed] Not just low latency, but volume too is key. That put quite a different spin on the point someone made that alpha opportunities have a life cycle and decay over time, as advantage is competed away, but may later return as robotrader focus moves on. This implies models have an ever-shorter half-life and need to be tuned to the current market regime and competitor learning curves. Clearly, this is anything but easy.

So it was scarcely surprising that artificial intelligence theories kept coming back into the conversation. At HIFREQ TRADE we heard more about Darwinian strategies for algo optimization as well as an intriguing reference to agent-based swarming strategies. Another quant stressed the need for a unified framework for idea generation and algo construction with back testing, simulation and production all tightly integrated. Algos should cycle between simulation and live trading, constantly adapting and evolving: it was indeed a very different world. The images were always from biological, not mechanical, engineering.

In this HFT world the discussion of risk was particularly important. It's not just a characteristic of the regime, someone insisted, but needs to be assessed relative to the strategy, the trader's risk appetite and technical capabilities. For a market neutral strategy the risks vary in real-time with the divergence from a flat position. Sudden regime shifts during moments of divergence can kill strategy returns. In a crisis non-correlated assets may suddenly synchronise their behaviour, creating unexpected exposures.

HIFREQ TRADE gave us much to ponder, but for me, perhaps, the choreography of large-scale basket trades and the stormy weather of trade winds sweeping across the market fabric were the most vivid images that will stay with me.