Professional Documents
Culture Documents
Forecasting in A Supply Chain
Forecasting in A Supply Chain
02/02/2021 2
Characteristics of forecasts
02/02/2021 3
Quantitative forecasting methods
Moving average
Exponential smoothing
Holt’s method
Winters’ method
Regressions
ARIMA/ARIMAX etc.
02/02/2021 4
Components of an observation
• Systematic component:
– Level (current deseasonalized value of a variable)
02/02/2021 6
Comparison of forecasting methods
02/02/2021 7
Moving averages
−1 −1+ −2 −2+ ⋯………………….+ − −
=
−1 + −2 + ⋯ … … … … … … … … . + −
Where:
n = no of periods
Note: for simple moving average, Wt-1 = Wt-2 = ….= Wt-n = 1 and for
weighted moving average ∑Wi = 1 (weights can be optimized)
02/02/2021 8
Exponential smoothing
+1 = + (1 − )
Where:
F1 = average (Di)
02/02/2021 9
Holt’s and Winters’ models
Systematic component = (level + trend)*seasonal factor
Where:
Intercept = level
Coefficient = trend
Holt’s method: only trend is addressed (two smoothing constants needed, α and
β, both varying between 0 and 1 (all can be optimized)
Winters’ method: both trend and seasonality are addressed (three smoothing
constants needed, α, β and γ, all varying between 0 and 1 (all can be optimized)
02/02/2021 10
Estimating seasonality
−1+( 2 )
−(2 )
+ +( 2 )
+∑ 2
+1−( 2 )
= ,
2
−1
+( )
2
= ( / ),
−1
−( 2 )
02/02/2021 11
ACF and white noise model
• Auto-correlation function(ACF):
02/02/2021 12
Tests for ACF and PACF
• Both tests the auto-correlations for the errors are different from
zero:
02/02/2021 13
ARIMA models
Consider a regression equation:
y = b0 + b1x1 + b2x2 +………………………….+bpxp + et
If we define:
x1 = yt-1, x2 = yt-2,………………………….., xp = yt-p
then:
yt = b0 + b1 y t-1 + b2 yt-2 +………………….+bp yt-p + et (AR)
and
yt = b0 + b1 e t-1 + b2 et-2 +………………….+bp et-p + et (MA)
02/02/2021 14
Box-Jenkins methodology
02/02/2021 15
Source: Box and Jenkins (1970) as said in Makridakis (1998)
ARIMA (p, d, q)
• AR: p = order of autoregressive part
02/02/2021 16
ARIMA models in general
ARMA model:
ARIMA model:
continued…..
02/02/2021 17
Continued…..
Seasonal ARIMA(1,1,1)(1,1,1)4:
02/02/2021 18
Identification of p, d and q
• Plot the data and observe to find out if there is trend and variance. If it
• Plot and the ACF (decides the MA part) and PACF (decides the AR
part).
• If the time plot is not horizontal or the ACF and PACF do not drop to
series.
02/02/2021 19
Non-stationary series
02/02/2021 20
ACF and PACF plots
02/02/2021 21
Data transformation
02/02/2021 22
ACF and PACF plots
02/02/2021 23
Estimating the parameters
02/02/2021 24
Measures of forecasting error
1. MSE (mean squared error) = (1/n)∑Et2
continued.....
02/02/2021 25
Continued.....
• Akaike’s information criterion:
Where:
• n = sample size
02/02/2021 26
Implications of forecasting error
02/02/2021 27
Forecasting in practice
02/02/2021 28
References/suggested readings
1. Box, G. E. P., Jenkins, G. M. (1970). Time series analysis: forecasting and control. San
Francisco: Holden-Day.
2. Chopra, S., & Meindl, P. (2012). Supply chain management: strategy, planning and
operation. New Delhi: Prentice Hall.
5. Simchi-Levi, D., Kaminsky, P., Simchi-Levi, E., & Shankar R. (2009). Designing and
managing the supply chain – concepts, strategies and case studies. New Delhi: Tata McGraw-
Hill Education.
02/02/2021 29
Disclaimer!
The content of this presentation is solely intended to be used by MBA-IV semester
“Supply Chain Analytics” students of School of Management Studies, University
of Hyderabad. Distribution of the same without the permission of the instructor is
strictly prohibited. The instructor, Dr. Pramod K. Mishra, declares that the sources
of various concepts/theories cited in the presentation are duly acknowledged.
Nobody should edit/post/upload any supply chain analytics (SCA) class materials
(cases, texts, ppts, video recordings or any other related contents) in/on to any
social networking sites, print media, YouTube etc. without the prior permission of
the concerned teacher “Dr. Pramod K. Mishra.” Disciplinary action, as per rule,
will be taken against the concerned student/s for such misconducts.
02/02/2021 30