CTL.

SC1x -Supply Chain & Logistics Fundamentals

Exponential Smoothing:
Level & Trend Data

MIT Center for

Transportation & Logistics
 Treatment of History
•  Previous models differed in the amount of history considered,
but were similar in their equal treatment of it.
n  Cumulative & Moving Average – equal weighting to all observations
n  Naïve – all weight to most recent observation
t t
∑ xi ∑ xi
x̂t ,t+1 = i=1
x̂t ,t+1 = i=t+1−M x̂t ,t+1 = xt
t M

•  Is there something in between these extremes?

•  Should we treat historical data differently?
n  The value of data degrades over time
n  Weight the newer observations more than the older ones
•  This is what exponential smoothing does
n  Each observation is weighted
n  Weights decrease exponentially as they age
(
x̂t ,t+1 = α xt + 1− α x̂t−1,t ) 0 ≤α ≤1

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 2
 Simple Exponential Smoothing
x̂t ,t+1 = α xt + 1− α x̂t−1,t ( ) 0 ≤α ≤1 Obs.   α=0.2   α=0.4   α=0.6  
t   0.2   0.4   0.6  
Recall that: t-­‐1   0.16   0.24   0.24  
t-­‐2   0.128   0.144   0.096  
x̂t−1,t = α xt−1 + 1− α x̂t−2,t−1 ( ) t-­‐3  
t-­‐4  
0.1024  
0.08192  
0.0864   0.0384  
0.05184   0.01536  
t-­‐5   0.065536   0.031104   0.006144  
Expanding and collecting terms: Weights attached to observations
( ( ) ) )(
x̂t ,t+1 = α xt + 1− α α xt−1 + 1− α x̂t−2,t−1 for different alpha values

2
x̂t ,t+1 = α x + α (1− α ) x + (1− α ) x̂
t t−1 t−2,t−1

Continuing to substitute:
2 3
( )
x̂t ,t+1 = α xt + α 1− α xt−1 + α 1− α xt−2 + 1− α x̂t−3,t−2 ( ) ( )
Which leads us to the general form:
0 1 2 3
( ) ( )
x̂t ,t+1 = α 1− α xt + α 1− α xt−1 + α 1− α xt−2 + α 1− α xt−3 . . . ( ) ( )
CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 3
 Exponential Smoothing Weights
1
(
x̂t ,t+1 = α xt + 1− α x̂t−1,t ) 0 ≤α ≤1
0.9

0.8

0.7 The alpha value indicates the value of “new” information

versus “old” information.
Effective Weight

0.6
As alpha approaches it limits:
0.5 •  αè1 leads to fast smoothing (nervous, volatile, naïve)
•  αè0 leads to slow smoothing (calm, staid, cumulative)
0.4

0.3

0.2

0.1

0
0 1 2 3 4 5 6 7 8 9 10
Age of Observation

Alpha=0.1 Alpha=0.3 Alpha=0.5 Alpha=0.7 Alpha=0.9

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 4
 Simple Exponential Smoothing

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 5
 Time Series Analysis
•  Simple Exponential Smoothing
n  Stationary demand – no trends or seasonality
n  Value of observations degrade over time
n  Utilizes a smoothing constant (α) where 0 ≤ α  ≤ 1
n  In practice 0.1 ≤ α  ≤ 0.3

Underlying  Model:  
We can also think of this as error-correcting.
xt    =    a    +    e  t
where:      
( )
x̂t ,t+1 = α xt + 1− α x̂t−1,t
x̂t ,t+1 = α xt + x̂t−1,t − α x̂t−1,t
e  t  ~  iid  (μ=0  ,  σ2=V[e])  
(
x̂t ,t+1 = x̂t−1,t + α xt − x̂t−1,t )
ForecasOng  Model:   x̂t ,t+1 = x̂t−1,t + α et
  x̂t ,t+1 = α xt + (1− α ) x̂t−1,t 0 ≤α ≤1 New estimate is the old estimate
plus some fraction of the most recent error.
 

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 6
 x^t,t+1
Example t
Exp. Smoothing
xt Alpha Alpha Alpha
Calculate the forecast for period 6 =0.7 =0.3 =0.1
from period 5 with alpha = 0.3: 1 109 109.0 109.0 109.0
2 92 97.1 103.9 107.3
( ) ( )
x̂5,6 = α x5 + 1− α x̂4,5 3 98
4 96
97.7
96.5
102.1
100.3
106.4
105.3
= (.3) (104) + (0.7) (100.3) = 101.4 5 104 101.8 101.4 105.2
6 98 99.1 100.4 104.5
What is the forecast for period 12 7 109 106.0 103.0 104.9
8 99 101.1 101.8 104.3
from period 10 with alpha = 0.3?
9 94 96.1 99.4 103.3
(hint: it is the same as the forecast for period 13, 14, . . .)
10 96 96.0 98.4 102.6
110
108
106
104
102
100
98
96
94
92
90
1 2 3 4 5 6 7 8 9 10 11

Actual Alpha=0.7 Alpha=0.3 Alpha=0.1

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 7
 Exponential Smoothing with Trend

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 8
 Time Series: Non-Stationary Models
x^t,t+1
170
Forecasts
Alpha = MA = 160

t xt 0.3 3 150
1 95 95.0
140
2 102 97.0
3 117 103.1 104.7 130
4 123 109.0 113.9
120
5 139 118.1 126.5
6 143 125.5 135.0 110

7 164 137.0 148.6 100

8 160 143.8 155.4
90
9 162 149.3 161.8
1 2 3 4 5 6 7 8 9 10 11
10 169 155.2 163.6
Actual Alpha=0.3 MA=3

Challenges:
•  Moving average & simple exponential smoothing
models will always lag a trend
•  They only look at history to find the stationary level
•  Need to capture the ‘trend’ or ‘seasonality’ factors
CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 9
 Time Series Analysis

Demand rate
a

•  Exponential Smoothing for Level & Trend time

n  Expand exponential smoothing to include trend
Often referred to as Holt’s Method

Demand rate
n 

n  Uses smoothing constants (α,β) where 0 ≤ α,  β  ≤ 1 b

Underlying Model: time

xt = a + bt + et
where: et ~ iid (µ=0 , σ2=V[e])
This is just x^t-1,t
Forecasting Model: x̂t ,t+τ = ât + τ b̂t
Updating Procedure:
( )(
ât = α xt + 1− α ât−1 + b̂t−1 ) The “old” trend -
estimated trend
These are estimates
from last period
of the level and
trend components
(
b̂t = β ât − ât−1 ) + (1− β ) b̂ t−1
at end of time
period t. The “new” trend - difference between this
CTL.SC1x - Supply Chain and Logistics Fundamentals period
Lesson: and
Exponential last forperiod’s
Smoothing estimated level.
Level & Trend 10
 Example Forecasting Model
x̂t ,t+τ = ât + τ b̂t
Suppose we are in time 101 Updating Procedure
and we use alpha=0.3 and ( )(
ât = α xt + 1− α ât−1 + b̂t−1 )
beta=0.1.
a)  Forecast demand for t=102
(
b̂t = β ât − ât−1 ) + (1− β ) b̂ t−1

b)  Forecast demand for t=110 Part a) Estimating demand for t=102
1. Find a^101
Data
( ) ( )( )
â101 = 0.3 x101 + 0.7 â100 + b̂100
t xt a^t b^t x^t,t+1
= (0.3) (95.0) + (0.7) (90.0 + 8.5) ≈ 97.5
100 92 90 8.5 98.5
2. Find b^101
101 95 97.5 8.4 105.9
( )( ) ( )
b̂101 = 0.1 â101 − â100 + 0.9 b̂100
Part b) Estimating demand for t=110
= (0.1) (97.5− 90.0) + (0.9) (8.5) ≈ 8.4
This means τ=9, so
x^101,110 =97.5+(9)8.4 = 173.1
3. Find x^101,102 =97.5+8.4 = 105.9
CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 11
 Example

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 12
 Example: #VMI1984
Spreadsheets are in resources
link for this video
You need to develop monthly forecasts (in pallets) for item #VMI1984
that seems to have an upward trend. Looking at past year’s data, you
have determined that α=0.25 and β= 0.10. Your estimated level (a^0) in
January (t=0) is 28 pallets/month and the estimate of trend (b^0) is 1.35.
a)  Using exponential smoothing, estimate a forecast for February
This is easy – just plug in the numbers.
x^J,F=a^J + (1)(b^J) = 28 + (1)(1.35) = 29.35 pallets
b)  It is now the end of February and demand was 27 pallets. What is
your forecast for March?

=D5+E5 =C6-F5

=$B$1*C6+(1-$B$1)*F5 =$B$2*(D6-D5)+(1-$B$2)*E5
CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 13
 Example: #VMI1984
Spreadsheets are in resources
link for this video
c) Build a spreadsheet for “next month” estimates for the next 8 months.

Actual Demand “Next Month” Forecasts

How good are the forecasts? Look at MSE=(1/n)Σe2, but which one?
We will need an estimate of the the forecast error for finding safety stock.
Keep a current running update of the MSE – using exponential smoothing. Select an
omega where 0.01≤ ω ≤0.1
2
(
MSEt = ω xt − x̂t−1,t ) + (1− ω ) MSE =$H$1*(C14-F13)^2+(1-$H$1)*I13
t−1
CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 14
 Damped Trends

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 15
 Damped Trends
•  Problems with trend terms
n  Trends do not continue unchanging indefinitely
n  Constant linear trends can lead to over-forecasting
n  This is especially true for longer forecast horizons
•  Damped trend model
n  Slight modification to exponential smoothing model
n  Select dampening parameter phi, where 0≤ ϕ ≤ 1
n  If ϕ = 1, this is just a linear trend

τ
Forecasting Model x̂t ,t+τ = ât + ∑ φ i b̂t
i=1

Updating ( )(
ât = α xt + 1− α ât−1 + φ b̂t−1 )
Procedure
(
b̂t = β ât − ât−1 ) + (1− β ) φb̂
t−1

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 16
 Same Example
Spreadsheets are in resources
link for this video
Build a spreadsheet for “next month” estimates for the next 8 months
using a damped trend.

=D6+$B$3*E6

=$B$1*C15+(1-$B$1)*(D14+$B$3*E14) =$B$2*(D15-D14)+(1-$B$2)*$B$3*E14

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 17
 Comparing Linear versus Damped
38
Next Month Forecasts Comparing  MSE:  
36        Linear:  4.52  
       Damped:    3.80  
Monthly Demand

34
 
32
Results  are  data  
30 specific,  obviously  
28 Actual Linear Damped

26
Jan Feb Mar Apr May Jun Jul Aug Sep Oct

42
Forecast @ EOM Jan
40
Nine  month  forecast   38
Monthly Demand

made  at  EOM  Jan.       36

  34
Note  the  tapering  effect   32
of  the  damped  model.     30

28

26
Feb Mar Apr May Jun Jul Aug Sep Oct

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 18
 Key Points from Lesson

CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 19
 Key Points
(
x̂t ,t+1 = α xt + 1− α x̂t−1,t )
•  Exponential Smoothing Models
n  Simple Model (level)
x̂t ,t+τ = ât + τ b̂t

n  Holt Model (level & trend) ( )(

ât = α xt + 1− α ât−1 + b̂t−1 )
(
b̂t = β ât − ât−1 ) + (1− β ) b̂ t−1

•  Other Smoothing Models

n  MSE Trending – for use in inventory models (ω)
n  Damped Trends – tapering effect (ϕ)
•  Core Concepts:
n  Value of information degrades over time
n  Balance of using both old & new information
CTL.SC1x - Supply Chain and Logistics Fundamentals Lesson: Exponential Smoothing for Level & Trend 20
 CTL.SC1x -Supply Chain & Logistics Fundamentals

Questions, Comments, Suggestions?

Use the Discussion!

“Lexi”
Photo courtesy Yankee Golden
Retriever Rescue (www.ygrr.org)
MIT Center for
Transportation & Logistics caplice@mit.edu