摘要：We start with formalizing the problem of assessing the stability of long-horizon forecasting of a given family of forecasting techniques. Next, we argue that in some applications the so-called singular-spectrum analysis (SSA) could be a suitable family of techniques to consider. We then apply the proposed methodology to two simulated time series and also to the Earth temperature records. We demonstrate that the SSA forecasts of the temperatures appear to be more stable at present than three-four years ago.
