Algorithms fed with huge amounts of data are increasingly capable of making smart predictions, even about such complex matters as the movement of people or the commercial performance of products.

In a move reminiscent of the movie Pi, a team from the University of Plymouth have developed an evolutionary algorithm that is being used to predict fluctuations in oil prices.

Smart algorithms

The algorithm, which was documented in a recently published paper, utilizes a range of programmed models to accurately predict the numerous rises and falls in oil price between 1986 and 2012.

Having been fed several years worth of data, the gene expression programming (GEP) model was capable of making almost perfect predictions about oil price performance, significantly outperforming more traditional methods.

Oil retains a hugely important role in the world economy, with previous price fluctuations causing significant economic issues around the world.

Being able to predict these fluctuations has thus far been a thankless task, due in no small part to the inherent complexity of the oil market, together with the various external factors that influence oil prices.

“The price of oil affects people everywhere, whether they live in countries that are net importers or exporters of the commodity. And the fluctuations of recent times have led to great economic uncertainty and that will only continue as consumption – and therefore demand – increased. If policy makers and economists had a tool which could accurately predict future prices, it would enable them to plan for the future at the same time allowing consumers to have an idea of the rising or falling costs they might incur,” the researchers say.

Cutting edge

GEP is at the leading edge of recent developments in artificial evolutionary systems, and has already shown its value in forecasting things such as exchange rates, electricity loads and various environmental measures.

The algorithm works via a tree structure that both learns and adapts to changing stimuli by changing its size and shape.  It’s designed to work much as a living organism does.  It has a simple genome, which together with a complex phenotype make it valuable for this kind of work.

Suffice to say, this is still very much at the margins of how forecasts are made, but if it keeps making successful predictions, it may make its way into the mainstream pretty rapidly.