See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/256088917

The Simple Exponential Smoothing Model

Conference Paper · September 2011

CITATIONS READS

65 33,337

2 authors:

Eva Ostertagova Oskar Ostertag

Technical University of Košice Technical University of Košice
39 PUBLICATIONS 1,806 CITATIONS 39 PUBLICATIONS 1,147 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Eva Ostertagova on 30 May 2014.

The user has requested enhancement of the downloaded file.

 MODELLING OF MECHANICAL AND MECHATRONIC SYSTEMS 2011
th
The 4 International conference
Faculty of Mechanical engineering, Technical university of Košice

THE SIMPLE EXPONENTIAL SMOOTHING MODEL

Eva Ostertagová1, Oskar Ostertag2

Abstract: In the paper there is prepared a theoretical base for calculation

and application of the simple exponential smoothing method. The simple
exponential smoothing model is one of the most popular forecasting methods
that we use to forecast the next period for a time series that have no
pronounced trend or seasonality.

Keywords: simple exponential smoothing model, forecast, smoothing

constant, root mean square error, MATLAB

1. Introduction
Exponential smoothing is probably the widely used class of procedures for smoothing
discrete time series in order to forecast the immediate future. This popularity can be attributed
to its simplicity, its computational efficiency, the ease of adjusting its responsiveness
to changes in the process being forecast, and its reasonable accuracy [3].
The idea of exponential smoothing is to smooth the original series the way the moving
average does and to use the smoothed series in forecasting future values of the variable
of interest. In exponential smoothing, however, we want to allow the more recent values
of the series to have greater influence on the forecast of future values than the more distant
observations.
Exponential smoothing is a simple and pragmatic approach to forecasting, whereby the
forecast is constructed from an exponentially weighted average of past observations. The largest
weight is given to the present observation, less weight to the immediately preceding
observation, even less weight to the observation before that, and so on (exponential decay
of influence of past data [1].

2. Forecasting with the simple exponential smoothing (SES) model

This forecasting method is most widely used of all forecasting techniques. It requires little
computation. This method is used when data pattern is approximately horizontal (i.e., there is
no neither cyclic variation nor pronounced trend in the historical data).
Let an observed time series be y1 , y2 ,K , yn . Formally, the simple exponential smoothing
equation takes the form of
yˆi +1 = α yi + (1 − α ) yˆi (1)
where yi is the actual, known series value for time period i, yˆi is the forecast value of the
variable Y for time period i, yˆi +1 is the forecast value for time period i + 1 and α is the

1
Eva Ostertagová, PhDr., PhD., Technical University of Košice, Faculty of Electrical Engineering
and Informatics, Department of Mathematics and Theoretical Informatics, Němcovej 32, 042 00 Košice, Slovak
republic. E-mail: eva.ostertagova@tuke.sk
2
Oskar Ostertag, doc., Ing. PhD., Technical University of Košice, Faculty of Mechanical Engineering, , Letná 9,
042 00 Košice, Slovak republic. E-mail: oskar.ostertag@tuke.sk

Modelling of Mechanical and Mechatronic systems 2011, September 20 – 22, 2011 Herľany, Slovak Republic 380
 MODELLING OF MECHANICAL AND MECHATRONIC SYSTEMS 2011
th
The 4 International conference
Faculty of Mechanical engineering, Technical university of Košice

smoothing constant [2]. The forecast yˆi +1 is based on weighting the most recent observation
yi with a weight α and weighting the most recent forecast yˆi with a weight of 1 − α .
To get started the algorithm, we need an initial forecast, an actual value and a smoothing
constant.
Since ŷ1 is not known, we can:

Set the first estimate equal to the first observation. Thus we can use ŷ1 = y1 .

Use the average of the first five or six observations for the initial smoothed value.
Smoothing constant α is a selected number between zero and one, 0 < α < 1 . When
α = 1 , the original and smoothed version of the series are identical. At the other extreme,
when α = 0 , the series is smoothed flat.
Rewriting the model (1) to see one of the neat things about the SES model
yˆi +1 − yˆi = α ( yi − yˆi ) , (2)
change in forecasting value is proportionate to the forecast error. That is
yˆi +1 = yˆi + α ei , (3)
where residual ei = yi − yˆi is forecast error for time period i. So, the exponential smoothing
forecast is the old forecast plus an adjustment for the error that occurred in the last forecast [1,
4].
By continuing to substitute previous forecasting values back to the starting point of the
data in model (1) we receive:
yˆi +1 = α yi + (1 − α ) [α yi −1 + (1 − α ) yˆi −1 ] = α yi + α (1 − α ) yi −1 + (1 − α ) 2 yˆi −1 ,
yˆi +1 = α yi + α (1 − α ) yi −1 + α (1 − α )2 yi − 2 + (1 − α )3 yˆi − 2 ,
yˆi +1 = α yi + α (1 − α ) yi −1 + α (1 − α ) 2 yi − 2 + α (1 − α )3 yi −3 + (1 − α )4 yˆi −3 ,
….
The forecast equation in general form is
yˆi +1 = α yi + α (1 − α ) yi −1 + α (1 − α ) 2 yi − 2 + L + α (1 − α )i − 2 y2 + α (1 − α )i −1 y1 =
i −1
= α ∑ (1 − α )k yi − k (4)
k =0

where yˆi +1 is the forecast value of the variable Y at time period i + 1 from knowledge of the
actual series values yi , yi −1 , yi − 2 and so on back in time to the first known value of the time
series, y1 [2, 3]. Therefore, yˆi +1 is the weighted moving average of all past observations.
The series of weights used in producing the forecast yˆi +1 is α , α (1 − α ), α (1 − α ) 2 , K .
These weights decline toward zero in an exponential fashion; thus, as we go back in the
series, each value has a smaller weight in terms of its effect on the forecast. The exponential
decline of the weights toward zero is evident [1].
After the model specified, its performance characteristics should be verified or validated
by comparison of its forecast with historical data for the process it was designed to forecast.
We can use the error measures such as MAPE (Mean absolute percentage error), MSE (Mean
square error) or RMSE (Root mean square error):

Modelling of Mechanical and Mechatronic systems 2011, September 20 – 22, 2011 Herľany, Slovak Republic 381
 MODELLING OF MECHANICAL AND MECHATRONIC SYSTEMS 2011
th
The 4 International conference
Faculty of Mechanical engineering, Technical university of Košice

1 n ei
MAPE = ∑ ⋅100 % , (5)
n i =1 yi
1 n 2
MSE = ∑ ei , (6)
n i =1
RMSE = MSE . (7)
Selection of an error measure has an important effect on the conclusions about which
of a set of forecasting methods is most accurate.
The speed at which the older responses are dampened (smoothed) is a function of the
value of α. When smoothing constant α is close to 1, dampening is quick and when α is close
to 0, dampening is slow [3]. If we want predictions to be stable and random variation
smoothed, use a small α. If we want a rapid response a larger α value is required.
Usually the MSE or RMSE can be used as the criterion for selecting an appropriate
smoothing constant. For instance, by assigning α values from 0.1 to 0.99, we select the value
that produces the smallest MSE or RMSE [1].

3. SES example with MATLAB

The simple exponential smoothing model can be illustrated by using data about number
of personnel in the industrial production in Slovakia over the years 2001 – 2010 that are in the
Table 1 below [5]:
Tab. 1 Number of personnel in industrial production in Slovakia over the years 2001 – 2010
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
486459 488240 493259 497302 518142 510698 526075 532853 447685 430658

We want to forecast the number of personnel in industrial production in Slovakia for the
year 2011.
By plotting of observed values we can see that we can use SES method. Since no forecast
is available for the first period, we will set the first estimate equal to the first observation.
For illustration here we try α = 0.4, α = 0.8 and α = 0.9.
The MATLAB script appears as [4]:
z=[486459,488240,493259,497302,518142,510698,526075,532853,447685,430658];
n=length(z);a=[0.4 0.8 0.9];
yi1=[];
yi1(1)=NaN;
yi1(2)=z(1);
for i=3:11
yi1(i)=a(1)*z(i-1)+(1-a(1))*yi1(i-1);
end
yi2=[];
yi2(1)=NaN;
yi2(2)=z(1);
for i=3:11
yi2(i)=a(2)*z(i-1)+(1-a(2))*yi2(i-1);
end
yi3=[];
yi3(1)=NaN;
yi3(2)=z(1);

Modelling of Mechanical and Mechatronic systems 2011, September 20 – 22, 2011 Herľany, Slovak Republic 382
 MODELLING OF MECHANICAL AND MECHATRONIC SYSTEMS 2011
th
The 4 International conference
Faculty of Mechanical engineering, Technical university of Košice

for i=3:11
yi3(i)=a(3)*z(i-1)+(1-a(3))*yi3(i-1);
end
RMSE1=...
sqrt((1/(n-1)*sum((z(2:end)-yi1(2:10)).^2)));
RMSE2=...
sqrt((1/(n-1)*sum((z(2:end)-yi2(2:10)).^2)));
RMSE3=...
sqrt((1/(n-1)*sum((z(2:end)-yi3(2:10)).^2)));
RM=[NaN,NaN,RMSE1,RMSE2,RMSE3];
i=1:11;
fprintf('\ti\t\tYi\t\t\tY^i(0.4) Y^i(0.8) Y^i(0.9)\n')
t=[i;z,NaN;yi1;yi2;yi3]';
t=[t;RM];
disp(t)
hold on
plot(i(1:10),z,'o:')
plot(i(2:11),yi1(2:11),'*g-')
plot(i(2:11),yi2(2:11),'+r-.')
plot(i(2:11),yi3(2:11),'xk--')
legend('actual values','forecast:0.4',...
'forecast:0.8','forecast:0.9')
hold off
Using SES with MATLAB we have received following data (see Table 2):
Tab. 2 Exponential smoothing forecast using MATLAB
Yi Y^i (0.4) Y^i (0.8) Y^i (0.9)
i
1.0 e+005*
1.0000 4.8646 NaN NaN NaN
2.0000 4.8824 4.8646 4.8646 4.8646
3.0000 4.9326 4.8717 4.8788 4.8806
4.0000 4.9730 4.8961 4.9218 4.9274
5.0000 5.1814 4.9268 4.9628 4.9685
6.0000 5.1070 5.0287 5.1377 5.1601
7.0000 5.2607 5.0600 5.1131 5.1123
8.0000 5.3285 5.1403 5.2312 5.2459
9.0000 4.4768 5.2156 5.3091 5.3203
10.0000 4.3066 4.9201 4.6433 4.5612
11.0000 NaN 4.6747 4.3739 4.3320
NaN NaN 0.3462 0.3148 0.3088
By calculating the smoothing values, we obtain the values presented in the columns 3 – 5
in the Table 2, for illustration by assigning α = 0.4, α = 0.8 and α = 0.9.
In the next to the last line in the Table 2 are forecast results for the time period 11,
therefore for year 2011. The resulting RMSE is in the last line in the Table 2.
So the forecast for year 2011 by using α = 0.9 is about 433200 personnel in the industrial
production in Slovakia.

Modelling of Mechanical and Mechatronic systems 2011, September 20 – 22, 2011 Herľany, Slovak Republic 383
 MODELLING OF MECHANICAL AND MECHATRONIC SYSTEMS 2011
th
The 4 International conference
Faculty of Mechanical engineering, Technical university of Košice

This concept is illustrated in Figure 1, which shows a time series observed for periods 1
to10 and the forecast for period 11.
5
x 10
5.4

5.2

4.8

4.6
actual values
forecast:0.4
4.4
forecast:0.8
forecast:0.9
4.2
1 2 3 4 5 6 7 8 9 10 11

Fig. 1 Plot for actual values and forecast values using α = 0.4, α = 0.8 and α = 0.9

4. Summary
With exponential smoothing the idea is that the most recent observations will usually
provide the best guide as to the future, so we want a weighting scheme that has decreasing
weights as the observations get older. The choice of the smoothing constant is important in
determining the operating characteristics of exponential smoothing. The smaller the value of
α, the slower the response. Larger values of α cause the smoothed value to react quickly − not
only to real changes but also random fluctuations [3]. Simple exponential smoothing model is
only good for non-seasonal patterns with approximately zero-trend and for short-term
forecasting because if we extend past the next period, the forecasted value for that period has
to be used as a surrogate for the actual demand for any forecast past the next period.
Consequently, there is no ability to add corrective information (the actual demand) and any
error grows exponentially.

Acknowledgement
The authors of this paper are grateful for financial support from GA VEGA, Project No.
1/0265/10 and this contribution is the result of the project implementation: Center for research
of control of technical, environmental and human risks for permanent development of
production and products in mechanical engineering (ITMS: 26220120060) supported by the
Research & Development Operational Programme funded by the ERDF.

References
[1] ACZEL, A. D., Complete Business Statistics, Irwin, 1989, ISBN 0-256-05710-8
[2] BROWN, R. G., MEYER, R. F., The fundamental theory of exponential smoothing, Operations
Research, 9, 1961, p. 673-685,
[3] MONTGOMERY, D. C., JOHNSON, L. A., GARDINER, J. S., Forecasting and Time Series
Analysis, McGraw-Hill, Inc., 1990, ISBN 0-07-042858-1
[4] OSTERTAGOVÁ, E., Applied Statistic (in Slovak), Elfa, Košice, 2011, ISBN 978-80-8086-
171-1
[5] Štatistický úrad Slovenskej republiky, URL: <http://www.statistics.sk>
Modelling of Mechanical and Mechatronic systems 2011, September 20 – 22, 2011 Herľany, Slovak Republic 384

View publication stats