Forecasting Week 8 – Forecasting Chapter 4

Document Sample
Forecasting Week 8 – Forecasting Chapter 4 Powered By Docstoc
					Week 8 – Forecasting
                (Chapter 4)


     Demand behavior, approaches to
     forecasting, measures of forecast
                   error


Rev. 03/16/10       SJSU Bus 140 - David Bentley   1
      Why Forecast?
   You’re wrong more than you’re right
   Often ignored or used as scapegoat
   Thankless job!




   Examples of the downside of forecasting


    Rev. 09/19/05   SJSU Bus 140 - David Bentley   2
   Why Forecast – (the answer)

We need to plan resources in advance!




 Rev. 09/19/05   SJSU Bus 140 - David Bentley   3
      Forecast accuracy
   Aggregation
        Would you rather forecast sales of all Ford
         automobiles or forecast a specific model?


   Time
        Would you rather forecast Ford sales for 2005
         or for 2010?


    Rev. 09/19/05      SJSU Bus 140 - David Bentley    4
        Forecast accuracy
   Aggregation
       Rather forecast sales of all Ford automobiles or
        forecast a specific model?
       Forecasts tend to be more accurate for
        groups of items than for individual items in the
        group
   Time
       Rather forecast Ford sales for 2005 or for 2010?
       Forecasts tend to be more accurate for the
        near future than for the distant future
    Rev. 09/19/05      SJSU Bus 140 - David Bentley   5
          Common Features of Forecasts
1.        Forecasts often (but not always) assume
          that what happened in the past will
          continue in the future
2.        Forecasts are rarely perfect
               “You are either lucky or lousy”
3.        Forecasts tend to be more for groups of
          items than for individual items
4.        Forecasts tend to be more accurate for
          the short range than for the long range
     09/25/06               SJSU Bus 140 - David Bentley   6
        Demand Components
   Components or Elements or Behavior
       Trend – long-term linear movement up or
        down
       Seasonal – short term recurring variations
       Cyclical – long-term recurring variations
       Random & Irregular – doesn’t fit other
        three components


    Rev. 02/12/02     SJSU Bus 140 - David Bentley   7
      Forecasting Approaches
    Qualitative (“subjective”)
         Judgment and Opinion
    Quantitative (“objective”)
         Associative
               External sources of data
         Historical
               Internal sources of data used


    Rev. 09/19/05           SJSU Bus 140 - David Bentley   8
      Judgment and Opinion - 1
   Sources
        Executives
        Marketing & Sales Projections
        Customers
        Potential customers
        “Experts”
              Delphi method


    Rev. 09/25/06         SJSU Bus 140 - David Bentley   9
      Judgment and Opinion - 2
    Appropriate Use
         Irregular or random demand
         New products
         Absence of historical data
    Techniques
         Surveys, questionnaires, interviews, focus
          groups, observation
         Delphi method

    Rev. 10/01/02      SJSU Bus 140 - David Bentley    10
      Associative
   Sources
        External industry data
        Demographic and econometric data
   Appropriate use
        Cyclical demand
   Technique
        Leading indicator, and
        Linear regression, in conjunction with
        Correlation
    Rev. 10/01/02     SJSU Bus 140 - David Bentley   11
      Historical
   Sources
        Historical (“time series”) data
   Appropriate use
        Varies (see later slides)
   Technique types
        Multi-period pattern projection
        Single period patternless projection

    Rev. 10/031/02      SJSU Bus 140 - David Bentley   12
     Multi-period Pattern Projection
     Techniques - Trend
   Appropriate use
        Clear trend pattern over time
   Techniques
        Best fit (“eyeball”)
        Linear trend equation or least squares
              Yt = a + bt
              b = n (ty) – (t)(y)
                      n t2 – (t)2
               a = y - b t
                         n


    Rev. 10/01/02            SJSU Bus 140 - David Bentley   13
        Multi-period Pattern Projection
        Techniques - Seasonal
   Appropriate use
       Seasonal demand
       Related to weather, holidays, sports, school
        calendar, day of the week, etc.
   Techniques
       Seasonal indexes or relatives
       Seasonally adjusted trend
             Separate trend from seasonality

    Rev. 10/01/02          SJSU Bus 140 - David Bentley   14
        Single Period Patternless
        Projection - 1
   Appropriate use
       Lack of clear data pattern
       Limited historical data
   Techniques
       Moving Average (older method)
             Ft = A
                    n
       Weighted moving average
             Ft = a(At-1) + b(At-2) + … + x(At-n)

    Rev. 10/01/02           SJSU Bus 140 - David Bentley   15
      Single Period Patternless
      Projection - 2
   Techniques (continued)
        Exponential Smoothing (newer method)
              Ft = Ft-1 + ( At-1 – Ft-1 )
        Naïve Forecast
              Simple (stable series)
                   = last period’s actual (often used with seasonality)
                   Ft = At-1
              Advanced (some trend)
                   Ft = At-1 + (At-1 - At-2)


    Rev. 03/13/07               SJSU Bus 140 - David Bentley               16
      Single Period Patternless
      Projection - 3
   Techniques (continued)
        Double exponential smoothing
              aka second order exponential smoothing
              Special case
              Incorporates some trend
              Uses exponential smoothing formula plus second
               formula with additional smoothing constant




    Rev. 10/01/02          SJSU Bus 140 - David Bentley         17
    Multiperiod Pattern Projection
   Behavior             Technique                             Tools

Trend             Trend line                          Linear regression or
                                                      Best fit (eyeball)
Seasonal          Seasonal calculations Seasonal relatives
                                                      (aka indexes)
Trend and         Seasonally adjusted                 Linear regression and
seasonal          trend                               seasonal relatives




  Rev. 09/25/06        SJSU Bus 140 - David Bentley                          18
    Single Period Patternless
    Projection
       Behavior         Technique                        Tools

Random and        Time series                     Moving average,
Irregular         (historical)                    weighted average,
                                                  exponential
                                                  smoothing, or
                                                  naïve
Random with some Time series                      Double exponential
trend                                             smoothing (aka trend
                 (historical)                     adjusted or second
                                                  order exponential
                                                  smoothing)

  Rev. 09/25/06    SJSU Bus 140 - David Bentley                       19
    Other Forecasting Methods
     Behavior         Technique                             Tools

Cyclical          Associative                      Leading indicator,
                                                   regression and
                                                   correlation
All behaviors     Judgment and                     Executive opinion, sales
                  opinion                          and marketing
                                                   estimates, and/or
                                                   customer surveys




  Rev. 09/25/06     SJSU Bus 140 - David Bentley                        20
      Measures of Forecast Error - 1
    Forecast Error (e, E, or FE)
         Et = At - Ft
    Average Error (AE)
         AE = E
               n
    Mean Absolute Deviation (MAD)
         MAD = |E|
                 n

    Rev. 09/25/06        SJSU Bus 140 - David Bentley   21
      Measures of Forecast Error - 2
    Mean Squared Error (MSE)
         MSE = E2
                n-1
    Standard Deviation (SD)
         SD = square root of E2
                              n-1
    Mean Absolute Percent Error (MAPE)
         MAPE = (|E|/A) (100)
                       n
    Rev. 09/25/06     SJSU Bus 140 - David Bentley   22
      Controlling the forecast - 1
   Control charts
       Upper and lower control limits
       (remember SPC?) – See Figure 3-11
   Formulas:                       ___
        Upper limit: 0 + z √MSE
        Lower limit: 0 – z √MSE
          where z = number of standard deviations
         from the mean

    Rev. 09/25/06     SJSU Bus 140 - David Bentley   23
      Controlling the forecast - 2
   Tracking Signal (TS)
        Reflects “bias” in the forecast
        TS = (A – F)
                    MAD
        Look for values within ± 4




    Rev. 09/25/06         SJSU Bus 140 - David Bentley   24
        Choosing and …
   Choosing a forecasting technique
       Nature of data (plot data: pattern?)
       Forecast horizon
       Preparation time
       Experience (may want to try several)
   Choosing a measure of forecast error
       Ease of use
       Cost

    Rev. 09/19/05     SJSU Bus 140 - David Bentley   25
      … Using
   Using forecast information
      Proactive vs. reactive
      Look at reasonability

      Assure everyone works off same data

      “What – if”




    Rev. 10/01/02   SJSU Bus 140 - David Bentley   26

				
DOCUMENT INFO