Methods for Forecasting Seasonal Items With Intermittent Demand
Demand Planning: Part 2
-
Upload
plato-larsen -
Category
Documents
-
view
25 -
download
0
description
Transcript of Demand Planning: Part 2
1
Demand Planning: Part 2
Collaboration requires shared information
2
Objectives Hands on experience using smoothing
procedures Enhanced trend and seasonal
smoothing models Forecasting into the future Parameter & initialization estimation
considerations Error measures
3
Moving Average modelFt+1 = Lt = (Dt + Dt-1 + ….+ Dt-n )/ n
Simple Exponential Smoothing
Ft+1 = Lt = α Dt + ( 1- α )Lt-1
Smoothing Models
Dt = sales in tLt = average in tFt = forecast in t
4
Forecasting Tools
Spreadsheets Example: Excel
install the Data Analysis Toolpack (Tools/Add-Inns/Analysis Toolpack)
open the file containing the data click on: Tools - Data Analysis (different options are
available) Other Add-ins: e.g., KADD and StatTools
Forecasting application software (2 types) statistical packages forecasting packages specifically designed for
forecasting applications
5
Hands on Exercise Hot Pizza exercise, problem 2, page 214
Use moving average (4 period) and simple exponential smoothing (alpha = .2 & .4) models with data, forecast weeks 13 to 16 into the future.
Northwestern Parts, (in class exercise for seasonal and tend enhanced models)
6
Components of demand
Trend component: growth or decline over an extended period of time
Cyclical component: wavelike fluctuation around the trend
Seasonal component: pattern of change that repeats itself year after year
Random component: after removal of other components
7
Pattern Issues
Which patterns are present in data? Things are not constant over time Need a process to identify change Need a procedure to update quickly
Enhancing Smoothing Procedures
8
Grow in SalesTrend Pattern
0
100
200
300
400
500
600
700
0 4 8 12 16 20 24 28 32 36
Quarter
Un
its
9
Update Equations with Trend:
Level: Lt = α ( Dt ) + ( 1- α ) ( Lt- 1+Tt-1 )
Trend: Tt = β (Lt - Lt-1 ) + ( 1- β ) Tt-1
Forecast Equation for ‘n’ period in the future:
Ft+n = Lt + n Tt
Expo with Trend - Update Equations and Forecasting Model
Dt = sales in tLt = average in tTt = trend in tFt = forecast in t
Basic Exponential Smoothing: Ft+1 = Lt = α Dt + ( 1- α )Lt-1
10
Trend Adjustment
Update smoothed average for recent trend Update Trend Factor
Difference of two period “Average” Weighted combination of
Past trend factor Current Forecast of trend factor
Trend the forecast
11
Seasonal Sale PatternSeason Pattern
0
100
200
300
400
500
600
700
0 4 8 12 16 20 24 28 32 36
Quarter
Un
its
12
Update Equations:
Level: Lt = α ( Dt / St ) + ( 1- α ) ( Lt- 1)
Season: St+p = γ ( Dt / Lt ) + ( 1- γ ) St
Forecast Equation for ‘n’ period in the future:
Ft+n = (Lt ) St+n
Expo with Season - Update Equations and Forecasting Model
Dt = sales in tLt = average in tSt = season in tFt = forecast in t p = season
Basic Exponential Smoothing: Ft+1 = Lt = α Dt + ( 1- α )Lt-1
13
Seasonality Adjustment Deseasonalize recent sales data Calculate smoothed average Update Seasonal Factor
Ratio of Actual to “Average” Weighted combination of
Past deseasonalized seasonal factor Current Forecast of seasonal factor
Seasonalize the forecast
14
Trend & Seasonality Common
Coca Cola Quarterly Sales in Millions of Dollars
$1,000
$1,500
$2,000
$2,500
$3,000
$3,500
$4,000
$4,500
$5,000
$5,500
15
Update Equations:
Level: Lt = α ( Dt / St ) + ( 1- α ) ( Lt- 1+Tt-1 )
Trend: Tt = β (Lt - Lt-1 ) + ( 1- β ) Tt-1
Season: St+p = γ ( Dt / Lt ) + ( 1- γ ) St
Forecast Equation for ‘n’ period in the future:
Ft+n = (Lt + n Tt ) St+n
Trend & Season (Winter’s) Update Equations and Forecasting Model
Dt = sales in tLt = average in tTt = trend in tSt = season in tFt = forecast in t p = season
16
Forecast Error Building a Forecast
Fit to historical data Project future data
Forecast Error How well does model fit historical data? Do we need to tune or refine the model? Can we offer confidence intervals about
our predictions?
17
Measuring Forecast Error MAD or MAE
average of the absolute errors Bias (tendency measurement)
Sum of all errors (plus & minus) MAPE (mean absolute percentage
error) Average absolute ratio of error to actual
MSE (mean square error) Square of all errors divided by ‘n’
18
Evaluating Forecast Models with Different Measures
Error in period t Mean Absolute Deviation Mean Absolute
Percentage Error Mean Squared Error
nfdMSE
nd
fdMAPE
nfdMAD
fde
tt
n
t
t
ttn
t
tt
n
t
ttt
2
1
1
1
)100(
19
Mabert Web Page Prepare: Specialty Packaging Corp. (A),
pp. 216-217. Develop forecasts for each quarter of 2007 for Clear and Black Plastic containers. Seasonal time series. Try using KADD analysis tool vs. provided Excel workbook.
Quiz: There will be a short quiz covering the fundamentals of demand planning and smoothing forecast models. Open book and notes.
20
Mabert Web Page
URL address with useful files:
http://kelley.iu.edu/mabert/class-e730.html