THE behaviour of complex systems such as financial markets may be far easier to predict and control than previously thought, suggests a new computer model of such systems.
Anything as messy as a financial market is usually considered too difficult to predict or manage. But physicist Neil Johnson and mathematician David Smith of the University of Oxford now claim that the prospects are not so bleak.
They explored a model designed to mimic a broad class of multi-component systems, from financial markets and ecosystems to complex computer software. In their model, 鈥渁gents鈥� within a population use various strategies to compete with one another for a 鈥渞esource鈥�. Simulating a financial market, for instance, involves thousands of such agents deciding whether to buy or sell stocks, basing their decisions on past trends in the prices of stocks.
Advertisement
To kick off the model, Johnson and Smith randomly equip their agents with strategies that represent the myriad ways in which real-life traders take decisions. They run the model using historical stock market data and compare the model鈥檚 predictions with the behaviour of the market. The model can be fine-tuned, by tweaking the strategies used by some agents for example, until the predictions fall in line with the market鈥檚 real behaviour.
Once tuned, the model is ready to make real predictions. Johnson has shown that the model accurately predicts stock market behaviour both hourly and by the minute ().
More striking is the possibility of steering markets in a desired direction. It turns out that just a few tweaks to the population of agents 鈥� for instance, altering the strategies used by a few agents, or taking a few agents out of the picture 鈥� can deflect the market onto a new trajectory. While this raises the spectre of powerful investors directing markets to their advantage, the model may also provide regulators with the insight to keep markets safe. 鈥淥ur work may suggest which trading practices should or shouldn鈥檛 be allowed,鈥� says Johnson.
The researchers stress that the model is applicable to all kinds of complex systems besides markets. For instance, they hope that this technique may ultimately be useful in medicine, with cells modelled as agents, to steer the biological system away from dangers such as tumours or epileptic fits.