Ballarin S.,Quantum Forecasting LLC |
Gervasi S.,Quantum Forecasting LLC |
Bacchetti S.,Quantum Forecasting LLC |
Capponi U.,Quantum Forecasting LLC |
And 8 more authors.
Neural, Parallel and Scientific Computations | Year: 2010
Physical motivations, theoretical aspects, and practical applications of a time-varying, auto-adaptive algorithm are described, as well as the results obtained through its application in some practical examples; these results were reached during a time span of over ten years from its first presentation. The intrinsic non-ergodicity of the physical phenomena leads us to hypothesize the existence of a characteristic time parameter, specific for each single physical phenomenon, uniquely valid in the temporal interval during which the same phenomenon is observed, in such a way as to transform the ergodic hypothesis into a locally valid ergodic approximation. The theoretical approach for determining the form of this time parameter springs from learning processes that take place without total memory loss. The algorithm's application to time series forecasts of any nature shows an extreme ease of utilization and an elevated forecasting capability, which vastly overcomes expected performances of forecasts obtainable through the use of tools derived from classical statistical methods. © Dynamic Publishers, Inc.