“Exponential smoothing methods have been
around since the 1950s, and are still the most popular forecasting methods used
in business and industry” (Rob J. Hyndman, 2008), which means that it’s
one of the most successful forecasting methods.
What is exponential smoothing?
Exponential smoothing (SE) is a model that
smooths the time series data using exponential purposes that gives weights reduced
over time(exponentially). It’s a popular method that creates a smoothed time
series. Exponential smoothing can be learned easily without a difficulty. It would
be better to be used if a new product has few sales data but then it will
provide a steady trend prediction of the future sales.
History of exponential smoothing:
There are a range of strategies that fall
under exponential smoothing, each one of these methods has the property of
forecasting a combination of weighted historical observations, with the latest
observations specified it will have a quite greater weight than observations at
the beginning. In addition, exponential smoothing gets to reflect that the
weights will get reduced exponentially when the observations become older.
In 1944 Robert G. Brown who at that time
was working for the US Navy as an operation research analyst, where he tracked
the speed and angle submarines firing using this method in a mechanical computing.
After that in the 1950s the method got extended to a discrete time series which
was a continuous time series at the beginning, it contained phrases to deal seasonality
At that time in 1956 presented one of his first applications
of forecasting the demand for spare parts the inventory system of the US Navy
in a meeting of the operations research society of America, then the methods
were further developed in 1963.
holt did work after that on a different exponential smoothing model than the
one Brown’s did with deference for the seasonality and trend way. His work was on
additive and multiplicative seasonal exponential smoothing in 1957. One of holt’s students Peter Winter wrote a
paper that provided empirical tests for his teacher method, the outcome of the
paper was called then Holt-Winters’ methods (or Winters).
1960, Holt’s method then become well recognized. In 1960 Holt did collaborate
with John Muth which he then introduced the first two long series of statistical
models. Exponential smoothing success in forecasting lead to several
researchers to find out models that can produce equivalent forecasts as these