Exponential Smoothing

Exponential smoothing, also called the exponential moving average, is a technique for smoothing time series data using an exponential window function. Unlike a simple moving average, which weights past observations equally, it applies exponentially decreasing weights over time, making it an easily learned and easily applied procedure that can incorporate assumptions such as seasonality, and it functions much like a low pass filter that removes high frequency noise from a series. Its basic recursive formula produces a smoothed output sequence from raw data using a smoothing factor between zero and one, so each smoothed value combines the current observation with the previous smoothed value. Simple exponential smoothing alone cannot forecast beyond the most recent data point, but double and triple exponential smoothing extend the method with trend and seasonal components that allow prediction. The technique traces back to Poisson's nineteenth century recursive exponential window functions and was developed further by Kolmogorov and Zurbenko in turbulence studies during the 1940s.