Escolar Documentos
Profissional Documentos
Cultura Documentos
Long Term
y 5+ years into the future y R&D, plant location, product planning y Principally judgement-based
Medium Term
y 1 season to 2 years y Aggregate planning, capacity planning, sales forecasts y Mixture of quantitative methods and judgement
Short Term
y 1 day to 1 year, less than 1 season y Demand forecasting, staffing levels, purchasing, inventory levels y Quantitative methods
Methods of Forecasting
y Naive Methods --- eye-balling the numbers; y Formal Methods --- systematically reduce forecasting errors; time series models (e.g. exponential smoothing); causal models (e.g. regression). y Focus here on Time Series Models
Forecasting Examples
Examples from student projects:
y y y y
Demand for tellers in a bank; Traffic on major communication switch; Demand for liquor in bar; Demand for frozen foods in local grocery warehouse.
70,000 items; 25 stocking locations; Store 3 years of data (63 million data points); Update forecasts monthly; 21 million forecast updates per year.
Ft 1 Ft 1
1 ! ( Dt Dt 1 . Dt 1n ) n 1 t D ! n i n i !t 1
Note that the n past observations are equally weighted. Issues with moving average forecasts:
y y y y
All n past observations treated equally; Observations older than n are not included at all; Requires that n past observations be retained; Problem when 1000's of items are being forecast.
...
today
Moving Average
Internet Unicycle Sales
450 400 350 300 250 200 150 100 50 0
Apr-01 Sep-02 Jan-04 May-05 Oct-06 Feb-08 Jul-09 Nov-10 Apr-12 Aug-13
n=3
Units
Month
Example:
Exponential Smoothing I
Include all past observations Weight recent observations much more heavily than very old observations:
weight Decreasing weight given to older observations
today
Exponential Smoothing I
Include all past observations Weight recent observations much more heavily than very old observations:
0 E 1
weight Decreasing weight given to older observations
today
Exponential Smoothing I
Include all past observations Weight recent observations much more heavily than very old observations:
0 E 1
weight Decreasing weight given to older observations
E E(1 E)
today
Exponential Smoothing I
Include all past observations Weight recent observations much more heavily than very old observations:
0 E 1
weight
E E(1 E) E(1 E) 2
today
0 E 1
weight
Ft ! aDt (1 a ) Ft 1
Ft k ! Ft
Exponential Smoothing
I
450 400 350 300 Units 250 200 150 100 50 0
Jan-03 May-04 Sep-05 Feb-07 Jun-08 Nov-09 Mar-11 Aug-12
(1000' )
E= 0.2
Month
Example:
Exponential Smoothing
Complicating Factors
Simple Exponential Smoothing works well with data that is moving sideways (stationary) Must be adapted for data series which exhibit a definite trend Must be further adapted for data series which exhibit seasonal patterns
Holts Method:
Actual Forecast
10
Month
Holts Method:
Bt ! EDt (1 E )( Bt 1 Tt 1 )
Smooth the trend forecast Tt
Tt ! F ( Bt Bt 1 ) (1 F )Tt 1
Forecast k periods into future Ft+k with base and trend
Ft k ! Bt kTt
ES with Trend
Inte net Unicycle ales (
450 400 350 300 Units 250 200 150 100 50 0
Jan-03 ay-04 Sep-05 Feb-07 Jun-08 No -09 ar-11 Aug-12
's)
Month
Example:
Winters Method:
Winters Method:
Dt Bt ! E (1 E )( Bt 1 Tt 1 ) St m
Smooth the trend forecast Tt
Tt ! F (
t 1
) (1 F )Tt 1
Dt St ! K (1 K ) St m Bt
Winters Method:
Ft k ! ( Bt 1 kTt 1 ) St k m
Smooth the trend forecast Tt
Dt St ! K (1 K ) St m Bt
's)
Month
Example:
Forecasting Performance
How good is the forecast? Mean Forecast Error (MFE or Bias): Measures average deviation of forecast from actuals. Mean Absolute Deviation (MAD): Measures average absolute deviation of forecast from actuals. Mean Absolute Percentage Error (MAPE): Measures absolute error as a percentage of the forecast. Standard Squared Error (MSE): Measures variance of forecast error
1 MSE ! n
(D
t !1
Ft )
Measures absolute error Positive and negative errors thus do not cancel out (as with MFE) Want MAD to be as small as possible No way to know if MAD error is large or small in relation to the actual data
MS
(D n
t!
Ft )
Measures squared forecast error -- error variance Recognizes that large errors are disproportionately more expensive than small errors But is not as easily interpreted as MAD, MAPE -- not as intuitive