Carl Carrie is global head of JP Morgan’s algorithmic products and Neovest, its execution management system subsidiary. He offered his views on the state of the art in algorithmic trading.
On the growing sophistication of algorithms
Traders today want more sophisticated and customizable algorithms–they now expect to be able to control how an algorithm responds intraday to price momentum, correlated sectors, the broad market, complex price-volume patterns or even news.
Rules are becoming much more complex. It used to be that you were just doing order-working. That meant you had fairly simple rules for how to slice an order up. Now, with so many choices in liquidity venues and so many different price points, the static rules-based mechanisms don’t really work as well. So you have to come up with a much richer, more dynamic and complex suite of logic. So, in a sense, you are often approximating the way a good human trader would act if he could think and react at microsecond speed.
On the trading parameters of algorithms
Traders can prime their algorithms with extra information. Do I want to limit the number of venues or relax some anti-gaming logic, or use more dark pool blocks, rather than smaller dark orders? Is today likely to be a special day in terms of liquidity, such as a FOMC day, or is implied price volatility unusually high? Sometimes, exceptional performance can be achieved by understanding the nuances of the way these algorithms perform.
On algorithms versus smart order routing
The short answer is that you need both tools–often on the same order. You need smart order routing to comply with best-ex obligations and to access liquidity in the most efficient manner possible for small lot sizes across displayed venues. That’s where smart order routers are typically highly efficient.
If you are trying to work 300 shares of Cisco, smart-order-routing technology will likely divvy that order up into three lots and find the right venue. But if you are trying to work 50 percent of the average daily volume of a very erratic stock such as MDRX with erratic volume and price patterns….It is precisely this sort of situation where smart routers are not as efficient as algorithms. Smart-order-routing technology knows a lot about current market conditions and small orders. It builds a book of sorts of all the liquidity it can see.
But when you are trying to work three times the average daily volume, there is a lot of liquidity to trade that the smart order router can’t see. That means you need much more sophisticated predictive and adaptive modeling techniques designed for larger orders. And that’s where an algorithm uses a lot more historical and contemporary knowledge about venues to place bets about where well-priced liquidity is likely to be.
There is a new science developing of how to keep the routers and algos balanced, which is sometimes tricky. You don’t want the smart order router to make all the decisions; otherwise you wouldn’t get a lot of blocks done. You don’t want an algorithm to make all the decisions; otherwise you’d make sub-optimal decisions with respect to small lots. Optimally, they should work together. They have to. You don’t want one dominating the other.
(c) 2008 Traders Magazine and SourceMedia, Inc. All Rights Reserved.
http://www.tradersmagazine.com http://www.sourcemedia.com
