The typical time series analysis technique which is utilized the occasion where in the future value is estimated from the time-series data. One of the weighted mean methods where among the past data which are obtained, you make big weight a newer data, extent small (it is past you decrease exponential function and) applying weight calculate moving average.

The latest occurrence has an influence on the occurrence immediately before strongly when and, when we would like to follow to the fluctuation of occurrence as much as possible and the like, we are suitable for the estimate of short term, for the order quantitative estimate in fixed time ordering system with such as stock management we are well used. In addition, it is used even with time series forecasting and stock price fluctuation analysis and the like with respect to business and financial affairs.

Simple exponential equity (primary system) arithmetic expression is displayed, like below.
Predicted value = a × Previous actual value + (1 - a) × previous predicted value
= Previous predicted value + a × (previous actual value - previous predicted value)

Namely, previous actual value which extent deviating from predicted value and calculating high, applying fixed factor a on that and allowing the correction value which can, in previous predicted value you deduce the latest predicted value. This time, as for actual value only is used in calculation of correction value with, as for this time becoming the base of predicted value the point where it is previous predicted value the point. Because of this, the fact that influence appears in the case where there is a unique actual value too much is avoided.

If there is a previous predicted value and a actual value, the simplicity which this time can calculate predicted value it is feature. However, previous predicted value because it is calculated from in the past time predicted value, the past data which is continued (predicted value) influence is little, but it remains.

Factor a (smooth constant) with, it sets in the range 0 < a < 1 that that past degree of influence of predicted value is decided. It seriously considers the value immediately before the extent whose a is close to 1, it means to seriously consider the extent past lapse which is close to 0. Usually with the past data which is accumulated doing, in order for prediction error of predicted value and actual value to become smallest, it sets the decision a, e.g., simulation it does.

Other than this simple index smooth method, the predicted value which is deduced with exponential smoothing (multi next smooth methods, heavily smooth method) there to be a secondary smooth method which it applies to another exponential smoothing as multi next systems, sesonality (periodicity) and the like the Portuguese method which considers tendency (Holt-Winters method) is famous. In addition, smooth factor constant do, there is also a variable response smooth method which takes in variable value from the time-series data, interweaves the tendency of actual value.