Can Predictive Algorithmic Switching be the Buy-Side’s Answer to High Frequency Trading?
Mon, 09 May 2011 08:24:00 GMT
An Interview with Marcus Hooper
In this interview for the High Frequency Trading Review, Mike O’Hara talks to Marcus Hooper, CEO & Executive Director, Pipeline Financial Group Limited (www.pipelinetrading.com). Marcus discusses Pipeline’s new Alpha ProSM service – which uses predictive algorithmic switching technology to enable buy-side firms to maximise access to liquidity and minimise adverse selection – and shares his thoughts on the changing nature of investors and speculators in today’s global equity markets.
High Frequency Trading Review: Marcus, can you start by giving us a brief intro to Pipeline, who you are and what you do?
Marcus Hooper: Certainly. Pipeline is a small company of around 100 people in total. We started in the US as an offshoot of NASDAQ’s complex trading group, when they were focusing on block trading and getting large order trading done. The product that was a consequence of that was the Block Board, which is a block matching system. In the US, it’s classed as an ATS (Alternative Trading System), here in Europe it’s an MTF (Multilateral Trading Facility).
HFTR: Is the Block Board a dark liquidity pool?
MH: No, it’s not fully lit but neither is it fully dark, which makes it fundamentally different. It uses a flagging system, which is effectively a heat map of trading scenarios that can be distributed to all of our clients, so they all see the same information. It has unique anti-gaming technology, which means you can’t just look at the heat map or enter an order, immediately derive information from either the heat map itself or how the order is handled, and exploit it. That’s the reason we can give it to everyone.
In effect, we’re obfuscating information, but we’re giving more than a dark pool would give. A dark pool would tell you nothing at all, apart from a trade print once you’ve executed. The consequence of that is that matching rates are very low in dark pools, well below 1%, whereas in our process we find that 25% match rates are about average.
HFTR: When you say you obfuscate information, does that mean you show that there is order interest on the book but not necessarily which side it is on?
MH: Correct. The starting point is what we call the orange light, which means there is available liquidity. But to trigger that orange light, orders going into the system have to qualify in certain ways.
The first qualification would be order size, because we have a minimum transaction size, different in the US to Europe, but the principle is the same, it has to be a meaningful quantity.
The next qualification is price. So if you put in an obviously untradeable limit price well outside the market reference price, then that would be disregarded and would not trigger an orange light.
As long as we’ve got an order of a certain size, it’s a firm order and it’s at or within the spread, that would trigger the orange light, which then tends to bring interest from other quarters.
As well as the anti-gaming technology, we have some randomization within the system, such that we wouldn’t instantly turn an orange light off if a price just moved slightly out of range. That’s because if the light keeps flashing on and off, it could be indicative of a limit order resting just outside the best bid or offer, so we delay the switching off of the orange light to prevent detection of such an order in that scenario.
HFTR: You recently launched something called Alpha Pro. What can you tell us about that?
MH: Alpha Pro is a recent development based around our Algorithm Switching Engine, which was a proprietary model we originally designed to limit and control information leakage.
It predicts the future performance of algorithms, by looking at a variety of things, such as the historical performance of a particular algorithm, the mechanism the algorithm is designed to support (e.g. VWAP or implementation shortfall), how aggressive or passive the algo is, etc. It then quantifies each and every algo it accesses based upon knowledge of what that algo is designed to do. It can also look at market conditions, for example if it’s a trending market, certain algos will perform better. Finally, it looks at user input, so if a user enters an order saying that he wants to be very aggressive for example, certain algos would be better than others.
The new prediction model is entirely proprietary, and allows the user to minimize market impact, given a particular scenario. You could think of Alpha Pro as a wrapper that sits around the switching engine. If you put an order into the engine, then based upon some very simple criteria (as outlined above), it would go off and execute your order to the best of its ability, which is all well and good.
However, there might be some other things about your order that are more complex. For example, if the system happens to know that you have a fund manager who always experiences a certain trading behaviour when he puts orders into the market because he’s getting stories from a certain analyst, the algo strategy can be customised specifically to fit that scenario.
HFTR: Can you explain how that might actually work in practice for a buy-side trader who is trying to adapt their trading strategies to counter the level of HFT activity in a particular market for example?