He found that forecasts fail because the models suffer from two problems : ( i ) they cannot capture the full detail of the underlying system, so rely on approximate equations; ( ii ) they are sensitive to small changes in the exact form of these equations.
