Brijesh Malkan
Traders Magazine Online News

Solving the Last Mile Problem in Investment Research

One executive takes a forward look at how will the research value proposition change over the short to medium term, and what are the products and strategies managers will turn to?

Traders Poll

Do you think it's a good idea to conduct an access fee pilot to assess the pricing models used by many trading venues?

Yes 0%
No 0%
Should have had a pilot program a long time ago. 100%


Free Site Registration

August 31, 2004

The Downside of Progress

By Doug Rivelli

Blind reliance on black-box trading algorithms is short-changing the buyside.

Sure, thanks to these mathematical wonders, commission levels are lower. Nevertheless, execution quality isn't better, liquidity access is still a problem, qualitative information doesn't flow any faster and reaction time to anomalous situations is reduced.

Why, then, is algorithmic execution gaining? The answer is that the broker dealers want you to use it because it's good for them. The more automated the trading conducted by the buyside, the less trading staff the sellside needs. The more streamlined they can become, the higher profit margins they can attain.

Many broker dealers have gone so far down the path of algorithmic trading, that there is no possible retreat. They have to make algorithmic trading work, or their existence is at risk.

The buyside, however, should not be lulled into believing that black-box algorithmic trading is the ultimate solution to execution problems. Algorithms are one tool. When algorithms are integrated with human insight, I believe clients receive optimal execution. But, used as a stand-alone solution, with no human oversight, algorithms are inadequate.

The nature of an algorithm is its biggest barrier to success. Algorithms are reactive. They need quantitative data, such as price or volume points, to make decisions. This is fine in a stock that is trading in its expected pattern. But when an unexpected event occurs, altering the expected trading pattern, the model can only react after the stock has started to react.

The result of this lag is lost opportunity and increased execution cost. A human, on the other hand, can immediately react, either maximizing trading profit or minimizing cost.

Algorithms also possess no ability to make a qualitative assessment of market situations. For example, what is the best way to react to a discount print, or should you sell into a hole? Models will answer these questions based either on pre-programmed responses or by analyzing purely quantitative data. What they cannot do is understand the context of the situation. Was it a clean up print? Was it an isolated market order through DOT that knocked the stock down? Was the specialist involved?

By answering these questions, a human trader can react in a more thoughtful manner than an algorithm. By analyzing a situation qualitatively, a human is less likely than a model to be trapped.

Stocks traded in black-box algorithms also cannot access upstairs liquidity. There is no knowledge of potentially interested counterparties. An algorithm cannot know if a buyer or seller has been lurking on the floor, does not know if there is more stock behind, does not know the source of the trade and cannot know if the specialist is ill and replaced by a weak substitute. All of this information can help a human trader take advantage of a situation, creating a strategy to optimize execution performance.

Lastly, an algorithm cannot dig itself out of a hole. If an execution managed by an algorithm is performing poorly, the algorithm does not have the ability to "rethink" its strategy and make adjustments significant enough to turn a losing trade into a winning trade. A human trader, however, can make a 180-degree strategy shift in mid-trade, can stop the balance of an order, or commit capital to rescue a trade.

However, algorithms can be a critical tool in a trader's arsenal. Quality algorithms are efficient at managing execution of stock trading in an "expected" manner, can allow a trader to systematically participate with price and volume and can allow a trader to efficiently manage large lists of stocks. But, to achieve best execution, humans must monitor these models, with the ability to interact with the orders and take control when necessary.

It is this integration of the mathematical algorithm and the human insight that can maximize execution performance. It is easy to drink the "Algorithm Kool-Aid" sold by many brokers, but you are likely to find that your best trades will have an equal measure of quant and human elements.

Doug Rivelli is senior vice president, program trading, Weeden & Co.