Abstract The forecast of dynamic processes is based on mathematical models
which usually describe determinate and stochastic characteristics of processes adequately.
Parameters values of considered models are estimated by observation data
over the processes in the past. It often occurs that there is data excess, which means
that additional number of data does not contribute to the precision of forecasting but
makes it more expensive. This work is devoted to the estimation of forecasting depth
at the fixed horizon. It was assumed that time series, of which forecast is considered,
possess some informative features that allow establishing the balance between the
horizon and the depth of the forecast. The algorithm of depth estimation has been
offered and real non-stationary time series has been analyzed. This analysis demonstrated
that there exists quasi-optimal depth for fixed horizon of forecasting.