EXSMOOTH

EXSMOOTH produces one period ahead forecasts for different models.

Notation
The following notation is used throughout this chapter unless otherwise stated:

Xt Observed series, t = 1, K , n

X$ t Forecast of one period ahead from time t

p Number of periods
k Number of complete cycles 2n p7
et 4
tth residual X t − X$ t −1 9
S0 Initial value for series

T0 Initial value for trend

I1− p , K , I 0 Initial values for seasonal factors

lp
ml Mean for the lth cycle, ∑ Xi p
1 6
i = l −1 p +1

Please note the following points:

• I1− p , K , I 0 are obtained from the SEASON procedure with MA = EQUAL if
p is even; otherwise MA = CENTERED is used for both multiplicative and
additive models.

• The index for the fitted series starts with zero.

• The value saved in the FIT variable for the tth case is X$ t −1 .

1
2 EXSMOOTH

Models
No Trend, No Seasonality Model

Xt = b + εt

Initial value

S0 = X

then

X$ 0 = S0 , e1 = X1 − X$ 0

St = St −1 + α et

X$ t = St

No Trend, Additive Seasonality Model

X t = b + It + ε t

Initial value

∑m i
i =1
S0 =
k
 EXSMOOTH 3

then

X$ 0 = S0 + I1− p

e1 = X1 − X$ 0

St = St −1 + α et

0 5
It = It − p + δ 1 − α et

X$ t = St + It − p +1

No Trend, Multiplicative Seasonality Model

X t = bI t + ε t

Initial value

∑m i
i =1
S0 =
k

then
4 EXSMOOTH

X$ 0 = S0 I1− p

e1 = X1 − X$ 0

St = St −1 + αet It − p

0 5
It = It − p + δ 1 − α et St

X$ t = St It − p +1

Linear Trend, No Seasonality Model

X t = b0 + b1t + ε t

Initial values

X n − X1
T0 =
n −1
1
S0 = X 1 − T0
2

then

X$ 0 = S0 + T0

e1 = X1 − X$ 0

St = St −1 + Tt −1 + α et

Tt = Tt −1 + αγ et

X$ t = St + Tt
 EXSMOOTH 5

Linear Trend, Additive Seasonality Model

X t = b0 + b1t + I t + ε t

Initial values

mk − m1
T0 =
0 5
k −1 p

p
S0 = X1 − T0
2

then

X$ 0 = S0 + T0 + I1− p

St = St −1 + Tt −1 + αet

Tt = Tt −1 + αγet

0 5
It = It − p + δ 1 − α et

X$ t = St + Tt + It − p +1

Linear Trend, Multiplicative Seasonality Model

X t = (b0 + b1t ) I t + ε t

Initial values

mk − m1
T0 =
0 5
k −1 p

p
S0 = m1 − T0
2
6 EXSMOOTH

then
X$ 0 = ( S0 + T0 ) I1− p

3
St = St −1 + Tt −1 + α et It − p 8
3
Tt = Tt −1 + α γ et It − p 8
0 51
It = It − p + δ 1 − α et / St 6
X$ t = ( St + Tt ) It − p +1

Exponential Trend, No Season Model

X t = b0b1t + ε t

Initial values

;
T0 = exp ln X2 − ln X1 = @ X2
X1

%& 1 X()
'
S0 = exp ln X1 −
2
ln T0 = 1
T0*
then

X$ 0 = S0 T0

St = St −1Tt −1 + αet

Tt = Tt −1 + αγet St −1

X$ t = St Tt
 EXSMOOTH 7

Exponential Trend, Additive Seasonal Model

X t = b0b1t + I t + ε t

Initial values

<1
T0 = exp ln m2 − ln m1 p 6 A
%& p ()
'
S0 = exp ln m1 −
2
ln T0
*
then

X$ 0 = S0 T0 + I1− p

St = St −1Tt −1 + αet

Tt = Tt −1 + αγet St −1

0 5
It = It − p + δ 1 − α et

X$ t = St Tt + It − p +1

Exponential Trend, Multiplicative Seasonality Model

X t = (b0b1t ) I t + ε t

Initial values

<1
T0 = exp ln m2 − ln m1 6 0k − 15A
%& p ()
'
S0 = exp ln m1 −
2
ln T0
*
then
8 EXSMOOTH

X$ 0 = ( S0 T0 ) I1− p

St = St −1Tt −1 + αet It − p

Tt = Tt −1 + αγet ( It − p St −1 )

0 5
It = It − p + δ 1 − α et St

X$ t = ( St Tt ) It − p +1

Damped Trend, No Seasonality Model

X t = b0 + φ b1t + ε t

Initial values

Xn − X1
T0 =
0 5
n −1 φ

1
S0 = X1 − T0
2

then

X$ 0 = S0 + φ T0

St = St −1 + φ Tt −1 + αet

Tt = φ Tt −1 + αγet

X$ t = St + φ Tt
 EXSMOOTH 9

Damped Trend, Additive Seasonality Model

X t = b0 + φ b1t + I t + ε t

Initial values

mk − m1
T0 =
0k − 15 pφ
p
S0 = m1 − T0
2

then

X$ 0 = S0 + φ T0 + I1− p

0
St = St −1 + φ Tt −1 + α 2 − α et5
1
Tt = φ Tt −1 + α α − φ + 1 et 6
It = It − p + δ 1 − α 02 − α 5 e
t

X$ t = St + φ Tt + It − p +1

Damped Trend, Multiplicative Seasonality Model

1 6
X t = b0 + b1φ t I t + ε t

Initial values

mk − m1
T0 =
0k − 15 pφ
p
S0 = m1 − T0φ
2
10 EXSMOOTH

then

1
X$ 0 = S0 + φ T0 I1− p 6
0 5
St = St −1 + φ Tt −1 + α 2 − α et It − p

1 6
Tt = φ Tt −1 + α α − φ + 1 et It − p

It = It − p + δ 1 − α 02 − α 5 e t St

1 6
X$ t = St + φ Tt It − p +1

References
Abraham, B., and Ledolter, J. 1983. Statistical methods of forecasting. New York:
John Wiley & Sons, Inc.

Gardner, E. S. 1985. Exponential smoothing: the state of the art. Journal of

Forecasting, 4: 1–28.

Ledolter, J., and Abraham, B. 1984. Some comments on the initialization of

exponential smoothing. Journal of Forecasting, 3: 79–84.

Makridakis, S., Wheelwright, S. C., and McGee, V. E. 1983. Forecasting: Methods

and Applications, 2nd ed. New York: John Wiley & Sons, Inc.