Kinetic equations for forecasting of non-stationary time-series

In this work the method of forecasting of non-stationary time-series with the use of kinetic equations for empiric distribution function is proposed. The series volume depends on the accuracy for any given horizon of forecasting. In dependence of statistics properties of so-called horizon series the evolution Liouville equation or Fokker-Planck equation are used. As an example the time-series of stock-exchange is considered.