It's a time series estimation method developed as an evolution of Exponential Smoothing (ES) and based on two different smoothing constants, and then on two fundamental equations:
The first function determines an estimate of the level Lt-1 as a weighted average of the observed series last instant of time available and the expected value at time t-2.
The second function determines the trend Tt-1 as a weighted average of two estimates: the last date update for the trend and that obtained by the difference of the last two estimated levels.
The smoothing constants α and β can be determined by minimizing the error measure of prediction.
Editor: Giuliano DI TOMMASO