There is a man-made system upon whose complex, poorly understood behaviour the lives of billions depend. This is no DNA-engineered super-virus, nor the simmering hydrocarbon gravy that is our atmosphere. It's the stock market.

Now research reported in the journal Physical Review Letters1 explains the statistics of stocks and shares in terms of the flocking behaviour of brokers on the market floor.

Trying to predict the way the market fluctuates has long been the goal of governments and financial analysts. While some countries bask in bright economic sunshine, others are thrown into turmoil -- all seemingly at the whim of chance.

The apparent randomness of the stock market is only skin-deep, as a glance at the statistics shows. Take for example the 'price return'. This is the difference between the purchase and the sale price of a share -- how much money is made or lost on a transaction.

If brokers simply decided on their own what and when to buy and sell, the distribution of returns -- how often returns of a given size are obtained -- would be 'Gaussian'. That is, exactly the spread you'd get from a random process like tossing a coin.

A closer look at the market statistics reveals that the return distribution is not 'Gaussian' but 'power law' -- the telltale mathematical sign that all is not random.

Researchers Victor Eguiluz and Martin Zimmermann, of the Mediterranean Institute for Advanced Study, Mallorca, Spain, wondered what would happen to the returns if, instead of acting alone, brokers followed herd instinct, just as sheep move en masse across a hillside, or as migrating wildebeest pour across the plains of Africa.

The 'herd' idea makes some sense. Brokers act on rumours that spread across the market floor. Big transactions trigger other brokers into action. Decisions are made on the say of computer software -- and brokers often use the same software.

Eguiluz and Zimmermann ran a computer simulation wherein 'agents' -- representing brokers -- could buy or sell, or wait. While waiting, agents made random links to other agents, building up a network. When one agent bought or sold, all the others in that agent's network did the same.

Summing up the calculated returns the researchers did indeed find a power law describing how often a return of a given size was made. This, they argue, shows that the power law in the market statistics could be explained by copycat behaviour amongst brokers.

Eguiluz and Zimmermann aren't alone in thinking cooperation between brokers is important. According to Thomas Lux of Kiel University, Germany, who has made more complex models of broker behaviour2, "this is one of about two dozen attempts to model financial markets."

But Geoff Rodgers at Brunel University, UK likes the simplicity of the herd model. "The best model is the one with the least details that best captures the observed features," he points out.

Analyst Jessica James at the First National Bank of Chicago thinks the agent-based approach could be especially useful in risk management -- estimating the chances of very large gains and losses. But, she cautions, "Different market rates have very different return characteristics. There is seldom enough data to be sure you have a model which agrees with the market."