Review: The danger of trading machines

October 20, 2012

By Martin Hutchinson

The author is a Reuters Breakingviews columnist. The opinions expressed are his own.

It’s a quarter-century since computerized program trading led to Black Monday on the U.S. stock market. A new and more advanced generation of arcane algorithms now threatens capital markets. That is the lesson of “Dark Pools,” a new book on machine-based equity trading by the Wall Street Journal’s Scott Patterson. The book is a great read – and raises an important question: could the trading machines destroy the capital markets?

Patterson provides a lively account of the growth of computerized stock trading, and of the market pools of undisclosed orders created by exploitative algorithms. The computerized trading pools have an inauspicious genesis, in a company run by a former trader at First Jersey Securities, the promoter of penny stocks which went bankrupt in 1987, its founder later going to prison for security fraud. Ethics, like many things, being path-dependent, First Jersey’s ethical standards appear to have permeated the computerized trading business as a whole.

Patterson explains the mechanics in sufficient detail for readers to work out why there’s a problem. It starts with the willingness of stock exchanges, first Nasdaq and now the NYSE, to pay for order-flow. Add in some clever tricks to manipulate the order queue and computerized traders can make a steady fee income from the exchanges, while most ordinary retail and institutional investors end up paying more.

Manipulating the order flow can be done by placing very low bids or very high offers for stock, which jump to the head of the order queue. When counterparties disappear as the computers cancel orders at superhuman speed, which often happens in turbulent markets, stocks can sell for one cent or $10,000 per share. Disappearing orders can be placed to test the market flow, and orders invisible to other participants in the market are also possible.

As yet, this high-frequency trading has not led to anything like 1987’s one-day 22 percent fall in the Dow Jones industrial average. But the occasional “flash crash” is alarming, as is the effect on large orders. These have become more expensive to execute, thanks to front-running computer-piranhas which move the price before big orders can be fully executed. Thus effective liquidity, the ability to buy and sell large institutional positions at close to the market price, has been greatly impeded – needless to say producing large additional profits for the algo traders.

In response, big orders are increasingly broken into smaller bits to avoid algo-detection. Patterson cites research, commissioned by the Financial Times, according to which average order size declined by 67 percent on the New York Stock Exchange and by 68 percent on Nasdaq between 2005 and 2010.

Patterson makes it clear that most algorithmic trading techniques rely on determining the order flow before competitors, thereby producing a “speed war”. This is basically a modern form of the “insider trading” practised by the bootleg-gin-sodden company director of 1929 who bought and sold based on prior knowledge of company results. In a fair “level playing-field” market, it would equally be impermissible.

Apart from the algo traders themselves, retail investors trading actively for their own account may have benefited from declining trading costs through the elimination of expensive and sluggish human traders and market-makers. However for individuals who trade only occasionally the benefit is minor and for the great majority, who invest through mutual funds or other managed investment arrangements, the losses appear to be substantial.

The book has many villains, some but not all of which have been driven out of business by the SEC, which makes valiant attempts to control the markets, but tends to be several years late. There are also a few heroes like Josh Levine, the unworldly programming genius who created Island, the first such pool. Patterson’s last chapter is devoted to a team who are undertaking long-term fundamental analysis through an artificial program, dubbed “Star”. They were close to unplugging Star when it began to pile into bombed-out, high-risk plays in the early months of 2009, but in the event the program was triumphantly vindicated. By avoiding the extremes of human euphoria and despair, Star and its successors may well prove useful.

Patterson makes it clear that this mess could be cleared up, or at least kept within bounds, but makes no recommendations as to how this should be done. Forbidding payment of fees for order-flow would clearly be one step. There is probably useful detailed work to be done in outlawing the methods that allow orders to be hidden or to disappear the moment they are needed. Above all, the market could be slowed by a “Tobin” transactions tax, at a very low rate of say 0.01 percent, which would quickly run up costs for the most abusive practices.

Could unconstrained fast trading crash the stock market altogether? Its scariest feature is that no one knows.

Post Your Comment

We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see http://blogs.reuters.com/fulldisclosure/2010/09/27/toward-a-more-thoughtful-conversation-on-stories/