One meaning may be:

You have data with noise. like foggy weather. then, you want to reduce fog and see things behind it. Smoothing may help you to reduce fog.

However, sometimes it is difficult to define what is noise and what is trend? When reducing noise, you may lose some underlying info.

For forecasting, smoothing may help you to extrapolate data. But, you implicitly assume that

- there is no meaningful causal relationships (or non-observable)

- there is only data generates itself

Then a simple rule-of-thumb method, smoothing, helps you find some estimations. It is simple, but it works well in most cases even better than most complicated methods.

Pablo Triana tells how mathematical complexity deteriorates financial system in his books. Similar perspective can be used for forecasting. Complexity does not ensure superiority.

Thanks for your answer Okan. I agree with you that in forecasting smoothing refers to a rejection of noise. But,don't you think that this meaning would go better with the word filtering instead, actually, exponential smoothing can be seen in the frequency domain as a low-pass filter, i.e., it allows to go trough low frequency components as trend, but they reject high frequency components that are associated to noise.

On the other hand, I totally agree with you that more complex models do not necessarilly means better forecasts.

Dear Juan,

your consideration about the similarity of smoothing and low-pass filtering is interesting.

Perhaps you can find some difference between this two concept if you look it from the following point of view.

Consider Winters' exponential smoothing.

I think that smoothing here is referred to the three sistematic demand components, i.e. level (L), trend (T) and seasonality coefficients(S), and not the overall demand 'signal'. This means that the variation of L, T and S are smoothed, and not the variation of demand.

For example, if data are supplied on a monthly based, and there is a seasonal demand pattern with period of one year, you have to 'smooth' separately L, T and S if you want to obtain forecasts that considers seasonality. This is very different than applying a low pass filter to the demand signal.

Inherent in the collection of data taken over time is some form of random variation. There exist methods for reducing of canceling the effect due to random variation. An often-used technique in industry is "smoothing". This technique, when properly applied, reveals more clearly the underlying trend, seasonal and cyclic components.

There are two distinct groups of smoothing methods :

1. Averaging Methods

2. Exponential Smoothing Methods