“The purpose is for traders to maximize their chances of getting a large block match within Pipeline,” says Fred Federspiel, Pipeline's president. Pipeline has a 20 percent fill rate, which is much higher than competitors, like Liquidnet which has a 4 percent match rate, says Federspiel. Pipeline is more effective at finding block matches because there’s no negotiation; it’s automatic execution, Pipeline's president notes, adding that Pipeline’s average trade size is 34,000 shares per trade as compared to dark pools like ITG POSIT whose average trade size is around 6,000 shares, he claims.

While buy-side traders are waiting in the block marketplace to achieve a block match, Pipeline’s proprietary Algorithmic Switching Engine, which uses real-time and historical data, “will predict what style of algorithm will be the most effective in the immediate future and it activates that style of algorithm for you and continually monitors the market for you,” Federspiel explains.

"If we can't get an order done in a block venue like Pipeline, we'll often utilize algorithms to complete the trade," stated David Brooks, director of Global Equity Trading at The Boston Company Asset Management LLC, in the release. Brooks said that Pipeline's "unique approach to automate switching between different algorithms makes sense for large blocks with longer time horizons, as well as small residual orders." Since the switching engine will work in tandem with the core Pipeline block matching functionality, "I'll get the best of both worlds," Brooks said.

Pipeline is assessing the different algorithms based on two fundamental issues: First, it’s looking at the cost side of being in the algorithm per block trade — which equates to the market impact that it incurs by participating in the market with a large order. Secondly, it’s looking at the rate at which shares are filled. In addition, it can compare the real-time trade prices that the algorithm achieves to nearby trades that other people are getting in the market.

If the algorithm is predicted to achieve poorly on any of those measures, then the algorithm switching engine will switch out of that algorithm to another one that will perform better in the next few minutes, he says. Typically the time frame it’s looking at is over the next 10 second to several minutes, he says.

However last month Pipeline expelled third party algorithms because they were sending most of their orders to lower-cost ECNs such as BATS Trading, which is 100 times cheaper per share than Pipeline, notes Federspiel. “The problem we had was customers were placing orders into an algorithm thinking the order was getting into Pipeline and it was not,” because the algorithm was going to a cheaper venue like BATS, he explains.

Though Pipeline eliminated algorithms as a distribution channel, it still has plenty of customers coming to it through third-party brokers, he explains. Goldman Sachs, Banc of America Securities and BNY ConvergEx are examples of brokers sponsoring DMA platforms that customers can use to reach Pipeline, he says.

Right now, Pipeline has 12 algorithms from bulge bracket providers and boutique algorithmic developers in testing, and several other providers have contacted Pipeline since Friday’s announcement, says Federspiel. “We’re just predicting the algorithm that will perform best in the immediate future,” he explains. “We’re providing best-of-breed access to participate while you maximize your chance within Pipeline of getting a giant match,” he says.

To create the Algorithm Switching Engine, Pipeline partnered with Adaptive Technologies, Inc., a leading provider of predictive analytics, co-founded by Pipeline’s Director of Research Henri Waelbroeck and with J. Doyne Farmer, Professor at the Santa Fe Institute, and co-founder of Prediction Company,, stated the release. Farmer is known for his research in the complex behavior of markets and modeling risk aversion, volatility and price formation, stated the release.

The initial deployment will occur toward the end of May, and it will simply allow traders to activate orders into the non-block markets, says Federspiel. Later, Pipeline will have a front-end which will accompany this Algorithm Switching Engine and it will allow people to see the routing decisions that are being made, says Federspiel. “In theory, people could go to some other algorithm that would accomplish the trading style that Pipeline’s trading engine is predicting to do well,” he says.

Pipeline is classifying algorithms into a handful of specific styles, such an algorithm that places pegged orders into reserve books. Another style is participation where an algorithm on an order-by-order basis decides to take liquidity or post liquidity. Another type is destination specific algorithms that focus on specific destinations or algorithms that avoid the NYSE, for example, he says. Then the prediction engine calculates which styles are likely to perform the best in the immediate future.