ch03- s3 - forecasting.pdf

Upload: batman92

Post on 12-Feb-2018

236 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    1/103

    Operations ManagementOperations Management

    Chapter 3Chapter 3

    Demand ForecastingDemand Forecasting

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    2/103

    OutlineOutline

    Where do we use forecasts

    Operations Management : Why do we need forecasts Forecastsdefinition

    Forecasting process Forecasting needs identification

    Horizon, period, aggregation level

    Data selection

    Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    2 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    3/103

    OutlineOutline

    Where do we use forecasts

    Operations Management : Why do we need forecasts Forecastsdefinition

    Forecasting process Forecasting needs identification

    Horizon, period, aggregation level

    Data selection

    Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    3 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    4/103

    ForecastForecast

    Forecast a statement about the future value

    of a variable of interest Forecasts are an important element in making

    informed decisions

    e ma e orecas s a ou suc ngs as wea er,demand, and resource availability

    4 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    5/103

    Where do we use Forecasts?Where do we use Forecasts?

    Weather

    Bettin on Racin

    Stock Market

    Operations Management and SC

    5 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    6/103

    Two Important Aspects of ForecastsTwo Important Aspects of Forecasts

    Expected level of demand

    The level of demand may be a function of somestructural variation such as trend or seasonalvariation

    Related to the potential size of forecast error

    6 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    7/103

    OutlineOutline

    Where do we use forecasts

    Operations Management : Why do we need forecasts Forecastsdefinition

    Forecasting process Forecasting needs identification

    Horizon, period, aggregation level

    Data selection Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    7 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    8/103

    Operations Management: Why do weOperations Management: Why do we

    need forecastsneed forecasts

    DistributionDecisions

    Distribution

    Decisions

    Assembling

    DecisionsManufacturing

    Decisions

    AssemblingDecisions

    TransportDecision

    Replenishment

    Decisions

    ManufacturingDecisions

    Man 6

    Ass 2

    Ass 3

    Ass 1Man 1

    Man 2 Man 4

    Man 3 Man 5

    Transp

    TranspSuppliers

    Demand

    Lead times

    8 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    9/103

    Why to Forecast in Operations?Why to Forecast in Operations?

    Supply Chain is in continuous change:

    Customers habits change Introduction of a new product

    Need to anticipate and to make decisions inadvance in order to optimize the supply chain

    Dr. Ali Cheaitou - 2013 9

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    10/103

    Customer DemandCustomer Demand

    Forecasting demand for finished goods or services isthe starting point for all operating activities in a

    Logistics System

    10 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    11/103

    Why to Forecast in Operations?Why to Forecast in Operations?

    Example 1

    Strategic selection of raw material supplier and of a3PL (third-party logistics provider) associated withthe production and distribution of a finished goodover several years

    e s ou ave, over t s p ann ng or zon an eaabout the demand:

    To ensure that the suppliers have adequatecapacities

    To be able to negotiate prices

    Dr. Ali Cheaitou - 2013 11

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    12/103

    Why to Forecast in Operations?Why to Forecast in Operations?

    Example 2

    Materials flow management of FMCG (Fast Moving Consumer

    Goods) Customers need to be delivered with very short lead times

    Retailer (supermarket) needs to:

    The manufacturer needs to:

    have raw material supplies

    produce a part or the totality of its production

    before receiving the customers orders.

    A need to have an idea about future demand

    Dr. Ali Cheaitou - 2013 12

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    13/103

    Why to Forecast in Operations?Why to Forecast in Operations?

    Production plans

    Personnel needs: full time, part time, contractual

    Capacity levels: equipment, machines, buildings

    urc ase requ remen s: raw ma er a s, componen s,services;

    Plans for subcontractor requirements

    Transport requirements: raw materials, finished

    goods and/or personnel

    Dr. Ali Cheaitou - 2013 13

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    14/103

    OutlineOutline

    Where do we use forecasts

    Operations Management : Why do we need forecasts Forecastsdefinition

    Forecasting process Forecasting needs identification

    Horizon, period, aggregation level

    Data selection Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    14 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    15/103

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    16/103

    Optimistic ForecastOptimistic Forecast

    Company estimates sales of

    6 million product unitsfor a particular year, but only sells

    4.5 million.

    What effect may this have on the business ? Excessive inventory of finished product

    Associated high storage costs

    Inventory becomes obsolescent Finished product must be sold at a loss

    Plant capacity is used unnecessarily

    Dr. Ali Cheaitou - 2013 16

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    17/103

    Pessimistic ForecastPessimistic Forecast

    Company estimates that it will sell

    6 million of its product units

    for a particular year

    Orders are received for 7.5 million.

    Results in inadequate stock (finsihed products) and lostorders poor customer relationships Insufficient raw material stops production

    Excessive costs due to subcontracting Excessive costs due to overtime/ to hiring of part time labour

    Dr. Ali Cheaitou - 2013 17

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    18/103

    Is it better to be Optimistic or Pessimistic?

    Pessimistic versus OptimisticPessimistic versus Optimistic

    Dr. Ali Cheaitou - 2013 18

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    19/103

    Pessimistic versus OptimisticPessimistic versus Optimistic

    It is considered better to be Optimistic rather than

    Pessimistic when forecasting.

    Pessimistic forecasts result in:

    oss o cus omers ue o na y o or ers

    demotivating effects on the staff (stress / moreproductive)

    high costs idle production line/labor due to

    insufficient raw material

    Dr. Ali Cheaitou - 2013 19

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    20/103

    Features Common to All ForecastsFeatures Common to All Forecasts

    1. Techniques assume some underlying causal systemthat existed in the past will persist into the future

    2. Forecasts are not perfect

    3. Forecasts for groups of items are more accurate thanthose for individual items

    4. Forecast accuracy decreases as the forecastinghorizon increases

    Dr. Ali Cheaitou - 2013 20

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    21/103

    Elements of a Good ForecastElements of a Good Forecast

    The forecast

    should be timely should be accurate

    should be reliable

    s ou e expresse n mean ng u un s

    should be in writing

    technique should be simple to understand and use

    should be cost effective

    Dr. Ali Cheaitou - 2013 21

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    22/103

    Forecast Accuracy and ControlForecast Accuracy and Control

    Forecasters want to minimize forecast errors

    It is nearly impossible to correctly forecast real-worldvariable values on a regular basis

    So, it is important to provide an indication of the

    value of the variable that actually occurs

    Forecast accuracy should be an importantforecasting technique selection criterion

    Error = Actual Forecast

    If errors fall beyond acceptable bounds, correctiveaction may be necessary

    22 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    23/103

    Forecast Accuracy MetricsForecast Accuracy Metrics

    n = tt ForecastActualMAD MAD weights all errors evenly

    n

    =

    100Actual

    ForecastActual

    MAPE t

    tt

    ( )2tt1

    ForecastActualMSE

    =

    n

    MSE weights errors according to theirsquared values

    MAPE weights errors according to

    relative error

    23

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    24/103

    PeriodActual

    (A)

    Forecast

    (F)

    (A-F)Error |Error| Error2 [|Error|/Actual]x100

    1 107 110 -3 3 9 2.80%

    2 125 121 4 4 16 3.20%

    3 115 112 3 3 9 2.61%

    Forecast Error CalculationForecast Error Calculation

    4 118 120 -2 2 4 1.69%

    5 108 109 1 1 1 0.93%

    Sum 13 39 11.23%

    n = 5 n-1 = 4 n = 5

    MAD MSE MAPE

    = 2.6 = 9.75 = 2.25%

    24 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    25/103

    OutlineOutline

    Where do we use forecasts

    Operations Management : Why do we need forecasts Forecastsdefinition

    Forecasting process Forecasting needs identification

    Horizon, period, aggregation level

    Data selection Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    25 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    26/103

    Forecasting ProcessForecasting Process

    The forecasting process objective is to determine:

    Future average value of a variable (demand, )

    Forecast accuracy (variability, measure of the difference between the average

    value and the different possible real values)

    Forecasting errors (high standard deviation) deteriorate theSupply Chain performance

    xamp e : sa es o

    with heat wave during summer

    without heat wave during summer

    Forecasting process should be improved in order to reduce theforecasts errors

    In the decision process, one must take into account the averagevalue and the error estimation

    26

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    27/103

    Forecasting Process: Forecasts definitionForecasting Process: Forecasts definition

    Three possible configurations:

    Forecasts users :salesman, producer, distributer,

    Forecasts experts (statistician)

    Forecasts committee: Users (marketing, sales): information collection, forecasts use

    Experts : models, statistical methods,

    Marketing Forecasts Sales Forecasts

    Statistical forecasts

    Forecasts Committee

    Final forecasts

    27

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    28/103

    Forecasting ProcessForecasting Process

    Six steps forecasting process:

    Forecasting needs identification

    Horizon, period, aggregation level Data selection

    Demand history (not sales history)

    romot ona campa gn,

    Information Systems

    Collaborative forecasts

    Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    28

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    29/103

    Forecasting ProcessForecasting Process

    Six steps forecasting process:

    Forecasting needs identification

    Horizon, period, aggregation level Data selection

    Demand history (not sales history)

    romot ona campa gn,

    Information Systems

    Collaborative forecasts

    Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    29

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    30/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative methods

    Based on experience, judgment, knowledge

    E.g. Expert Judgement, Data gathering methods, etc..

    Quantitative methods

    Based on data, statistics

    E.g. Causal methods, Extrapolative methods (timeseries)

    30

    Dr. Ali Cheaitou - 2013

    F i PF i P

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    31/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative methods

    Based on experience, judgment, knowledge

    E.g. Expert Judgement, Data gathering methods, etc..

    Quantitative methods

    Based on data, statistics

    E.g. Causal methods, Extrapolative methods

    31

    Dr. Ali Cheaitou - 2013

    F ti PF ti P

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    32/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative methods

    When do we use Judgemental Methods

    When little information about sales history is available.

    This ha ens for exam le when the com an is launchin

    a new product, and needs to know the likely demand

    32

    Dr. Ali Cheaitou - 2013

    F ti PF ti P

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    33/103

    Qualitative methods

    Expert Judgement

    It is normal to start by asking someone who is an expert in thearea to make an estimation which may be subjective not based on a quantitative models

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    33

    Dr. Ali Cheaitou - 2013

    F ti PF ti P

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    34/103

    Qualitative methods

    Help can be provided to the expert as additional information, this canbe done using different Data Gathering methods

    Market Research:

    Assess the demand for a new product by survey

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Historical Comparisons:

    Review available information about similar products or processes

    Scenario Analysis:

    In case of judgemental forecasting associated with uncertainty:

    consider and analyse the range of all possible outcomes

    34

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    35/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative methods

    Disadvantages

    Very numerous demands to be forecast very high workload

    Many factors could disrupt the forecaster (being in a bad or good

    Difficulty in determining the forecast errors (due to the difficulty indetermining the knowledge degree of the expert)

    35

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    36/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative methods

    Improvement : Delphi method

    Ask experts their opinion by anonymous forms and separately (inorder to avoid the inter-influence between experts)

    First statistical results (based on first answers) are sent to newexper s

    The new experts formulate their own forecasts:

    knowing the first results

    must justify their choice if it is very different from the first one

    By iteration

    either Consensus

    or General Opinion + Some Divergence

    36

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    37/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative methods

    Based on experience, judgment, knowledge

    E.g. Expert Judgement, Data gathering methods, etc..

    Quantitative methods

    37

    Dr. Ali Cheaitou - 2013

    Based on data, statistics E.g. Causal methods, Extrapolative methods

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    38/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Quantitative methods

    Explicit mathematical models + historical data

    Can be automated by computer programs Can be analyzed and improved more simply than the qualitative methods

    Two sub-families

    a s ca ex rapo a ve me o s : s or ca a a ynam ca evo u ons a

    are extrapolated for the future periods

    Causal methods (statistical regressions): historical data relationshipbetween the forecast variables(demand) and one (or different ) parameter(s).

    Example: sales of mineral water and the external temperature

    38

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    39/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Qualitative - Quantitative methods

    The quantitative and qualitative methods are not exclusive of eachother.

    Use a combination of different methods based on:

    The existence of historical data with different natures

    rs y, use e s or ca eman a a w an ex rapo a ve me o orecas

    software) secondly, include in this first model the subjective opinion of the experts

    concerning for example a promotional campaign

    39 Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    40/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Quantitative methods

    Extrapolative methods (Time series) Long lifecycle products averaging methods (arithmetic average, movingaverage, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    40 Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    41/103

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

    Quantitative methods

    Extrapolative methods (Time series) Long lifecycle products averaging methods (arithmetic average, movingaverage, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    41 Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    42/103

    g ppg ppTime SeriesTime Series

    Time Series : series of time-ordered sequence of observations with aconstant frequency (periodicity)

    Dt: demand in period t, t=1,2,, T

    DT

    D

    (mean of D)

    D(standard deviation)

    Time Series Forecasts: Forecasts that project patterns identified inrecent time-series observations, by assuming that future values of thetime-series can be estimated from past values of the time-series

    T

    t

    D

    =

    =

    1

    T

    DT

    t

    Dt

    D

    =

    =1

    2)(

    42 Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    43/103

    g ppg ppTime SeriesTime Series

    D

    = 5229.25

    D= 3215.94

    Low D data quasi constant simple forecasting process

    High D (20% of D) fluctuating data difficult forecasting process

    43 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    44/103

    TimeTime--SeriesSeries BehaviorsBehaviors

    Trend: long term upward or downward movement in the data (related topopulation shift, changing income)

    Seasonality: short term, fairly regular variations (related to calendar,events, time of the day)

    economic or political changes)

    Irregular variations: due to unusual circumstances (weather conditions,strikes,). Their inclusion in the series may distort overall picture. Shouldbe identified and removed.

    Random variation: residual variations that remain after all otherbehaviors have been accounted for.

    44 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    45/103

    TimeTime--Series BehaviorsSeries Behaviors

    45 Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    46/103

    ggForecasting MethodsForecasting Methods

    Quantitative methods

    Extrapolative methods (Time series) Long lifecycle products averaging methods (arithmetic average, moving

    average, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    46 Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    47/103

    Forecasting Approaches and MethodsForecasting Approaches and Methods

    Ft: demand forecast value in period t, t=1,,T,

    for 1t T Ft: simulation of the realized demand

    using the forecasting modelfor T

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    48/103

    They do not take account of external factors but

    they look at a series of past values to predictwhat will happen in the future

    Extrapolative MethodsExtrapolative Methods

    There are many different kinds of extrapolativemethods: Nave method (Arithmetic average) Simple Moving average

    Weighted moving average Exponential smoothing

    48 Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    49/103

    Forecasting Approaches and MethodsForecasting Approaches and MethodsNave methodNave method

    This approach assumes that:

    data from the immediately preceding past can be used

    to forecast needs for the next period

    DT

    t

    Can be used with

    a stable time series (historical data from period to periodshowing changes are insignificant )

    Trend

    TF tt

    ==

    49 Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    50/103

    Example

    Given the following sales history for 5 weeks, Calculatethe forecasts sales value for week 6, using the simpleavera e method?

    Forecasting Approaches and MethodsForecasting Approaches and MethodsNave methodNave method

    Answer : 5230

    Week 1 2 3 4 5 6

    ProductSales

    8996 4531 2362 4249 6012

    50 Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    51/103

    g ppg ppTimeTime--Series ForecastingSeries Forecasting -- AveragingAveraging

    These Techniques work best when a seriestends to vary about an average Averaging techniques smooth variations in the data

    They can handle step changes or gradual changesin the level of a series

    Techniques1. Straight moving average2. Weighted moving average

    3. Exponential smoothing

    Dr. Ali Cheaitou - 201351

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    52/103

    Technique that averages a number of the mostrecent actual values in generating a forecast

    Moving AverageMoving Average

    1==

    =

    An

    i

    it

    averagemovingin theperiodsofNumber

    1periodinvalueActualaveragemovingperiodMA

    periodfor timeForecast

    where

    1

    =

    =

    =

    =

    n

    tAn

    tF

    n

    t

    n

    t

    nt

    Straight Moving AveragesStraight Moving Averages

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    53/103

    Straight Moving AveragesStraight Moving Averages

    Historical data from the npast time periods are used to forecast futureactivity

    Assume that past sales of a particular product were January12500, February 16000, March 13500, April 11500, May

    14000. If n =3, then for June the forecast will be

    13500 + 11500 + 14000= 13000

    When June sales results are available July forecast can becalculated, based on April, May and June sales.

    The choice of n should be optimized by : Calculating the error between Dt and Ft for all tT

    Choosing n that minimizes this error

    M value influences the forecasts values: Low n values keeps the deterministic fluctuations

    High n values eliminates the random fluctuations

    53 Dr. Ali Cheaitou - 2013

    St i ht M i ASt i ht M i A

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    54/103

    Straight Moving AveragesStraight Moving Averages

    Example

    Using the same data as for the previous example,

    calculate the forecasts sales value for week 6, usingthe moving average method, with n=4?

    Answer : 4289

    Week 1 2 3 4 5 6

    ProductSales

    8996 4531 2362 4249 6012

    54 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    55/103

    Weighted Moving AveragesWeighted Moving Averages

    Weighted moving average models apply

    Weighting to period data.

    and considers

    Some periods more important than others.

    with

    and

    TnDFt

    nti

    iit =

    =

    ,1

    10 i 11

    =

    =

    n

    nti

    i

    55 Dr. Ali Cheaitou - 2013

    ExampleExample

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    56/103

    Weighted Moving AveragesWeighted Moving Averages

    Assume figures for past sales of a product are as shownin the following table

    From experience the following weights are applied 1=0.5 on April 2= 0.3 on March 3= 0.2 on February

    a s e orecas or ay =

    Sales over four month period

    January 10000

    February 13500

    March 11500

    April 14000

    56

    Dr. Ali Cheaitou - 2013

    ExampleExample

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    57/103

    Weighted Moving AveragesWeighted Moving Averages

    The forecast for May is

    (0.2 x 13500 + 0.3 x 11500 + 0.5 x 14000) = 13150

    the greater is the smoothing effect.

    57 Dr. Ali Cheaitou - 2013

    Exponential SmoothingExponential Smoothing

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    58/103

    Models are developed using a single weighting or

    smoothing factor, named

    is between 0 and 1 but never achieves values of 0or 1

    The mathematical model is Ft+1 = Dt + (1- ) Ft

    F = Estimated sales (Forecast)D = Actual sales (realized Demand)

    = smoothing factor

    58 Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    59/103

    Why is it called Exponential Smoothing ?Why is it called Exponential Smoothing ?

    weight

    10

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    60/103

    Why is it called Exponential Smoothing ?Why is it called Exponential Smoothing ?

    Ft+1 = Dt + (1- ) Dt-1+ (1- )2 Dt-2 ++ (1- )

    t D0

    When 1 : the model increases the importance of the more

    more reactive in case of change

    is optimized by:

    Performing a simulation over the historical data (t=1T)

    Taking the value of that minimizes the difference between Dt and Ftfor all t=1T

    60 Dr. Ali Cheaitou - 2013

    Exponential SmoothingExponential Smoothing

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    61/103

    Exponential SmoothingExponential Smoothing

    ExampleExample

    Motor car dealer predicted sales of 23000 vehicles in

    March

    Actual sales were 22150

    factor is 0.40

    Calculate forecast sales for April

    61 Dr. Ali Cheaitou - 2013

    Exponential SmoothingExponential SmoothingE l C l l tiE l C l l ti

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    62/103

    Example CalculationExample Calculation

    Forecast sales for April is

    FApril = DMarch + (1- )FMarch

    = 0.4x22150 + 1-0.4 x23000

    = 22660 vehicles

    62 Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    63/103

    Forecasting MethodsForecasting Methods

    Quantitative methods

    Extrapolative methods (Time series) Long lifecycle products averaging methods (arithmetic average, moving

    average, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    63 Dr. Ali Cheaitou - 2013

    Techniques for TrendTechniques for Trend

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    64/103

    Techniques for TrendTechniques for Trend

    Linear trend equation

    Non-linear trends

    64 Dr. Ali Cheaitou - 2013

    Methods for Historical DataMethods for Historical DataPresenting a Linear TrendPresenting a Linear Trend

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    65/103

    Presenting a Linear TrendPresenting a Linear Trend

    The trend is in general a long term effect of certain parameters

    65

    Dr. Ali Cheaitou - 2013

    Methods for Historical DataMethods for Historical DataPresenting a Linear TrendPresenting a Linear Trend

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    66/103

    Presenting a Linear TrendPresenting a Linear Trend

    The trend is in general a long term effect of certain parameters

    Ft=a+bt

    where tis the timevariable

    and

    aand barecalculated byminimizing the error

    66

    Dr. Ali Cheaitou - 2013

    Methods for Historical DataMethods for Historical DataPresenting a TrendPresenting a Trend

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    67/103

    To forecast future values of the demand (Ft), values of aand bareestimated based on previously acquired data Dt, t=1,T

    They are calculated , such as to minimize the error

    where Dt is the given historical data and Ft=a + btT

    FDT

    t

    tt=

    1

    2)(

    Presenting a TrendPresenting a Trend

    The obtained values are given by:

    and2

    11

    2

    1 1 1

    =

    ==

    = = =

    T

    t

    T

    t

    T

    t

    T

    t

    T

    t

    tt

    ttT

    DttDT

    b

    T

    tbD

    a

    T

    t

    T

    t

    t = =

    =1 1

    67

    Dr. Ali Cheaitou - 2013

    Methods for Historical DataMethods for Historical DataPresenting a TrendPresenting a Trend

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    68/103

    Presenting a TrendPresenting a Trend

    68

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    69/103

    gg

    Quantitative methods

    Extrapolative methods (Time series) Long lifecycle products averaging methods (arithmetic average, moving

    average, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    69

    Dr. Ali Cheaitou - 2013

    Techniques for SeasonalityTechniques for Seasonality

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    70/103

    Techniques for SeasonalityTechniques for Seasonality

    Seasonality regularly repeating movements in seriesvalues that can be tied to recurring events

    Expressed in terms of the amount that actual valuesdeviate from the average value of a series

    Models of seasonality

    - Seasonality is expressed as a quantity that gets added to orsubtracted from the time-series average in order toincorporate seasonality

    Multiplicative

    - Seasonality is expressed as a percentage of the average (ortrend) amount which is then used to multiply the value of a

    series in order to incorporate seasonality

    70

    Dr. Ali Cheaitou - 2013

    Models of SeasonalityModels of Seasonality

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    71/103

    Models of SeasonalityModels of Seasonality

    71

    Dr. Ali Cheaitou - 2013

    Multiplicative Method for Historical DataMultiplicative Method for Historical DataPresenting a Trend and a SeasonalityPresenting a Trend and a Seasonality

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    72/103

    ese t g a e d a d a Seaso a tyese t g a e d a d a Seaso a ty

    If the data present a seasonality, then the seasonality factor should becalculated and taken into consideration for the different seasons (highseason/low season for example).

    Cycle (high + low seasons) = T periods High season = from period i to period j, then the seasonality factor Ch(high season) and Cl (low season) are given by:

    The forecasts from the linear regression (trend) are then adjusted usingthese coefficient.

    =

    ==

    T

    t

    t

    itt

    h

    D

    DC

    1

    =

    = +=

    +

    =T

    t

    t

    t jttt

    l

    D

    DD

    C

    1

    1 1

    72

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    73/103

    Quantitative methods

    Extrapolative methods (Time series) Long lifecycle products averaging methods (arithmetic average, moving

    average, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    73

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    74/103

    Example: fashion products, PCs, CDs, books,

    Innovative products

    Manufacturer (retailer) must decide about the quantities to beproduced (ordered) before knowing the exact future demand

    The need a realistic estimation of the forecasts error

    74

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    75/103

    Short life-cycle products have an important demand variability

    In general, for short life cycle products we do not have an importanthistorical data (demand)

    Time series models are suitable for long life-cycle products

    Solution: a combination between a good production and inventorymanagement with a partial anticipation & a short term forecasts

    75

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    76/103

    Forecasting methods

    A qualitative A priori forecast realized by experts, long time before the sellingseason

    Error: in general around 50-100%

    After observation of the first periods realized demand quantitative (extrapolative)estimation

    76

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:Forecasting MethodsForecasting Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    77/103

    Quantitative methods

    Extrapolative methods (Time series)

    Long lifecycle products averaging methods (arithmetic average, movingaverage, exponential smoothing)

    Long lifecycle products - Trend

    -

    Short lifecycle products

    Causal methods (regressions)

    77

    Dr. Ali Cheaitou - 2013

    Causal MethodsCausal Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    78/103

    Causal methods are based on a known or perceived relationship betweenthe factor to be forecast and other external or internal factors

    Knowledge of variables used in forecast development:

    Quantity to be forecast is the dependant variable Other variables are independent

    Define t, =

    Ft, t>T : estimated (forecast) of the demand

    Simple linear regression assumes a linear relationship exists between thedependent variable Ft, and a single independent variable zt. Therelationship may be expressed as

    To forecast future values of the demand (Ft), values of aand bareestimated based on previously acquired data Dt, t=1,T

    Ft = a + bzt

    78

    Dr. Ali Cheaitou - 2013

    Causal MethodsCausal Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    79/103

    To forecast future values of the demand (Ft), values of aand bareestimated based on previously acquired data Dt, t=1,T

    They are calculated , such as to minimize the error

    where Dt is the given historical data and Ft=a + bztT

    FDT

    t

    tt=

    1

    2)(

    The obtained values are given by:

    and2

    11

    2

    1 1 1

    =

    ==

    = = =

    T

    t

    t

    T

    t

    t

    T

    t

    T

    t

    T

    t

    tttt

    zzT

    DzDzT

    b

    T

    zbD

    a

    T

    t

    T

    t

    tt = =

    =1 1

    79

    Dr. Ali Cheaitou - 2013

    Forecasting ProcessForecasting Process

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    80/103

    Six steps forecasting process:

    Forecasting needs identification

    Horizon, period, aggregation level

    Data selection

    Demand history (not sales history)

    romot ona campa gn,

    Information Systems

    Collaborative forecasts

    Forecasting method choice

    Make the forecast

    Forecasts errors analysis

    80

    Dr. Ali Cheaitou - 2013

    Forecasting Process:Forecasting Process:Forecasts errors analysisForecasts errors analysis

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    81/103

    Forecasting models :

    Should fit the real world scenario as closely as possible

    Do not provide perfect predictions

    A model with wide discrepancies is invalid and should be revised orcompletely rebuilt from start

    An error analysis permits to measure the forecasting process performance

    improve (if necessary) the whole (or a part) of the forecasting process

    To test model accuracy apply

    Mean Absolute Deviation (MAD)

    81

    Dr. Ali Cheaitou - 2013

    Mean Absolute DeviationMean Absolute Deviation(MAD)(MAD)

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    82/103

    MAD is derived from the absolute deviation

    This is the absolute (positive) value of the forecast error

    Forecast error: the difference between actual value Dtand forecast value Ft in the same time period

    Absolute deviation = |Dt-Ft|

    Mean absolute deviation (MAD) =

    For each developed model MAD is calculated. The modelwith lowest MAD value should be the preferred model.

    T

    FDT

    t

    tt=

    1

    82

    Dr. Ali Cheaitou - 2013

    Forecasting Approaches and MethodsForecasting Approaches and MethodsError measurementError measurement

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    83/103

    Other measurements can be used to estimate the forecast error:

    Simple difference: et=Dt-Ft

    Mean (algebraic) error

    ( )FDT

    t

    tt=

    1

    Mean squared error (deviation)

    T

    FDT

    t

    tt=

    1

    2)(

    83

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    84/103

    Numerical Application

    84

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    85/103

    85

    Dr. Ali Cheaitou - 2013

    Causal MethodsCausal Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    86/103

    Causal methods are based on a known or perceived relationshipbetween the factor to be forecast and other external or internal factors

    Knowledge of variables used in forecast development:

    Quantity to be forecast is the dependant variable Other variables are independent

    Define Dt, t=1 T : realized value of the demand

    Ft, t>T : estimated (forecast) of the demand

    Simple linear regression assumes a linear relationship exists between thedependent variable Ft, and a single independent variable zt. Therelationship may be expressed as

    Ft = a + bzt

    86

    Dr. Ali Cheaitou - 2013

    Causal MethodsCausal Methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    87/103

    To forecast future values of the demand (Ft), values of aand bareestimated based on previously acquired data Dt, t=1,T

    They are calculated , such as to minimize the error

    where Dt is the given historical data and Ft=a + bztT

    FDT

    t

    tt=

    1

    2)(

    The obtained values are given by:

    and2

    11

    2

    1 1 1

    =

    ==

    = = =

    T

    t

    t

    T

    t

    t

    T

    t

    T

    t

    T

    t

    tttt

    zzT

    DzDzT

    b

    T

    zbD

    a

    T

    t

    T

    t

    tt = =

    =1 1

    87

    Dr. Ali Cheaitou - 2013

    Example of Causal MethodExample of Causal Method

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    88/103

    A company manufactures a successful range of products,

    it wants to:

    Forecast product demand for the next year, knowing that :

    - Product sales are affected by the advertising budget.

    - Sales and advertising budgets for 8 quarters (2 years) areshown in the following table. Product advertising/salesrelationships is shown in the following chart.

    88

    Dr. Ali Cheaitou - 2013

    T bl f E l F tT bl f E l F t

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    89/103

    Table for Example ForecastTable for Example Forecast

    Quarter Advertising

    budget

    Sales

    1 15.0 153

    2 17.5 198

    Product Sales

    200

    250

    3 12.0 147

    4 8.5 104

    5 9.5 131

    6 12.5 159

    7 14.5 1608 11.0 124

    0

    50

    100

    150

    0 5 10 15 20Advertising

    S

    ales

    89

    Dr. Ali Cheaitou - 2013

    Calculate the Sales ForecastCalculate the Sales Forecast

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    90/103

    The budget for advertising in quarters 9, 10,

    11 and 12 is 12.0, 17.0. 20.0 and 14.0 respectively.

    9,10,11 and 12 ?

    90

    Dr. Ali Cheaitou - 2013

    Calculate the values forCalculate the values for

    Quarters 9 10 11 & 12Quarters 9 10 11 & 12

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    91/103

    Quarters 9, 10, 11 & 12Quarters 9, 10, 11 & 12Quarter Advertising

    budget

    Sales

    1 15.0 153

    2 17.5 198

    3 12.0 147

    .

    5 9.5 131

    6 12.5 159

    7 14.5 160

    8 11.0 124

    9 12.0

    10 17.0

    11 20.0

    12 14.091

    Dr. Ali Cheaitou - 2013

    RegressionRegression

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    92/103

    Straight line in chart is the regression line calculated tominimise the sum of squares of the error terms, that is to

    minimise

    8

    ((Predicted sales in uarter i) (actual sales in uarter i))2

    If Dt is the sales in quarter tand zt is the advertising budgetin quarter t, regression equations are used to calculate the

    best fit line. The values obtained are displayed on equationin the chart.

    i = 1

    92

    Dr. Ali Cheaitou - 2013

    Forecast: Example ChartForecast: Example Chart

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    93/103

    Forecast: Example ChartForecast: Example Chart

    Product Sales

    200

    250

    t= . t .

    0

    50

    100

    150

    0 5 10 15 20

    Advertising

    Sa

    les

    Sales

    Linear (Sales)

    93

    Dr. Ali Cheaitou - 2013

    Calculation ExampleCalculation Example

    Values obtained are displayed in the equation on the chart

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    94/103

    Values obtained are displayed in the equation, on the chart,

    a = 36.74and

    b = 8.78,

    so that yt , the forecast sales quarter is given by

    Ft = 8.7773z + 36.735.

    This relationship can be used to forecast the sales in the next

    four quarters on the basis of the budgeted advertising in thesequarters as follows

    Quarter 9 = 8.78 x 12.0 + 36.74 = 142

    Quarter 10 = 8.78 x 17.0+ 36.74 = 186 Quarter 11 = 8.78 x 20.0+ 36.74 = 211.94 Quarter 12 = 8.78 x 14.0+ 36.74 = 159.66

    94

    Dr. Ali Cheaitou - 2013

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    95/103

    95

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    96/103

    Example: fashion products, PCs, CDs, books,

    Innovative products

    Manufacturer (retailer) must decide about the quantities to beproduced (ordered) before knowing the exact future demand

    The need a realistic estimation of the forecasts error

    96

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    Short life-cycle products have an important demand variability

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    97/103

    S o t e cyc e p oducts a e a po ta t de a d a ab ty

    In general, for short life cycle products we do not have an importanthistorical data (demand)

    Time series models are suitable for long life-cycle products

    Solution: a combination between a good production and inventorymanagement with a partial anticipation & a short term forecasts

    97

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    Forecasting methods

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    98/103

    g

    A qualitative A priori forecast realized by experts, long time before the sellingseason

    Error: in general around 50-100%

    After observation of the first periods realized demand quantitative (extrapolative)estimation

    98

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    N : number of products of the same products family

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    99/103

    p p y

    T: periods of the product (family) life cycle

    Di,t: demand of product i at period t

    All products have the same demand shape

    Tobs: observation horizon length (number of time periods)Tobs

    : ratio of the demand during Tobs over thewhole demand of product i

    : in order to get a standard criteria (valid for

    all products)

    =

    ==

    T

    1t

    ti,

    1t

    ti,

    obsci

    D)(TF

    N

    Tobs==

    N

    1i

    ci

    obsc

    )(F

    )(TF

    99

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    The forecast for product i over all the life cycle:

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    100/103

    )(

    D

    F

    Tobs

    1t

    ti,

    1ti,

    obsc

    T

    t TF

    =

    =

    =

    For any product of the same family: if we observe a total demand of

    then the whole demand forecast (for all the life cycle) will be

    )(

    DobsT

    1t

    ti,

    1

    ,

    obsc

    T

    t

    tiTF

    F =

    =

    =

    =

    obs

    T

    1t

    ti,D

    100

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    Numerical application

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    101/103

    pp

    N = 3 products

    Observation horizon Tobs = 2 periods

    Life cycle of all the products T = 8 periods

    Period Product A Product B Product C

    For a product D, of the same products family as products A, B and C, if ademand of 5200 products is observed over 2 periods, what will be the totaldemand forecast over the entire life cycle (8 periods)?

    2 1850 1800 1850

    3 2300 2800 1950

    4 2500 2250 2250

    5 1200 2100 2450

    6 1000 1700 2500

    7 900 1300 2000

    8 700 1000 1300

    101

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    Numerical application

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    102/103

    0.26811750

    3150

    D

    D

    )(TF:AProductT

    1ttA,

    Tobs

    1t

    tA,

    obscA ===

    =

    =

    DTobs

    0.241

    14700D

    )(TF:BProductT

    1t

    tB,

    1t

    ,

    obscB ===

    =

    =

    0.22416050

    3600

    D

    D

    )(TF:CProduct T

    1t

    tC,

    Tobs

    1t

    tC,

    obscC ===

    =

    =

    102

    Dr. Ali Cheaitou - 2013

    Extrapolative Methods: Short LifeExtrapolative Methods: Short LifeCycle ProductsCycle Products

    Numerical application

  • 7/23/2019 CH03- S3 - Forecasting.pdf

    103/103

    244.00.2240.2410.268

    3

    )(F)(F)(F)(F

    )(TFcCcBcA

    3

    1i

    ci

    obsc

    =++

    =

    ++

    ==

    = obsobsobs

    obs TTT

    N

    T

    For product D, the estimated demand over the 8-period selling season is:

    21311244.0

    5200)(

    D2

    1t

    tD,8

    1

    , ===

    =

    = obsct

    tDTF

    F

    103

    Dr. Ali Cheaitou - 2013