Ratan Jha (21 MAY 2008) Advisor: Dr. Larry Lapide · In some scenarios application of simple...

Preview:

Citation preview

Ratan Jha

(21 MAY 2008)

Advisor: Dr. Larry Lapide

AgendaBackground

Product Hierarchy

Statistical Forecasting

Results

Recommendations

Jha 21 May 2008 2

Jha 21 May 2008 3

Players: Wireless Inc (Buyer) and Hitec Inc (Seller)

Process: Both collaborate to match demand with supply generate

mutually agreed upon consensus forecast or collaborative forecast

Author's last name(s) XX May 2008 4

How should collaborative forecast information

be used for mitigating demand uncertainty?

AgendaBackground

Product Hierarchy

Statistical Forecasting

Results

Recommendations

Jha 21 May 2008 5

Jha 21 May 2008 6

Family A

Assembl

yBoard CableBase Memor

yRouter Power Family B

Assembl

yBoard FeatureBase Power Cable Switch

Product Family Level

Product Type Level

PID (Product ID) Level

PIDs

PIDs

Jha 21 May 2008 7

AgendaBackground

Product Hierarchy

Statistical Forecasting

Results

Recommendations

Jha 21 May 2008 8

Jha 21 May 2008 9

Periods of Zero Demand between Periods of non-Zero Demand at

PID Level. Also called Intermittent or Erratic Demand Pattern.

Jha 21 May 2008 10

Negligible Periods of Zero Demand at Product Type Level Termed

Continuous Demand (Arbitrary Naming).

Jha 21 May 2008 11

Damped trend model for continuous

demand

20–Period Moving Average for

intermittent demand

Forecast generated both at type level

and PID level

Bottom Up and Top Down forecasting

approach

Root Mean Squared Error(RMSE) as

measure of accuracy

1 2 3 10

Initialization

Parameter Fitting

11 12 13 24

25 26 27 42

Forecasting

Jha 21 May 2008 12

Product Type

PID1 PID2 PID3 PID4

10 8 15 19

52 –

RMSE:12

Forecast At Product Type Level = 56 (RMSE: 10)

Forecast at PID Level

Bottom-Up Approach

Product Type

PID1 PID2 PID3 PID4

56

11 9 16 20

Top Down Approach

Jha 21 May 2008 13

Collaborative forecast at PID level in monthly buckets available

PID level statistical forecast aggregated into monthly buckets

A Hybrid forecast weighted by statistical and collaborative forecast

in any given time period is generated using following equation:

Hybrid Forecast = α* Statistical Forecast + (1-α)*Collaborative Forecast,

where α

is a parameter estimated using optimization techniques

Collaborative forecast at PID level in monthly buckets available

PID level statistical forecast aggregated into monthly buckets

A Hybrid forecast weighted by statistical and collaborative forecast

in any given time period is generated using following equation:

Hybrid Forecast = α* Statistical Forecast + (1-α)*Collaborative Forecast,

where α

is a parameter estimated using optimization techniques

AgendaBackground

Product Hierarchy

Statistical Forecasting

Results

Recommendations

Jha 21 May 2008 14

Jha 21 May 2008 15

Collaborative Forecastworst among the three

Jha 21 May 2008 16

PID Collaborative RMSE Statistical RMSE Hybrid RMSE1 2.52 5.92 2.52 (α

= 0)2 9.33 7.39 7.39 (α

= 1)3 15.64 11.76 11.31 (α

= 0.77)4 36.18 11.43 11.43 (α

= 1)5 46.74 7.87 7.87 (α

= 1)6 339.89 65.39 26.48 (α

= 0.93)7 1.15 3.11 0.90 (α

= 0.20)8 2.52 4.36 2.52 (α

= 0)9 4.24 1.00 1.00 (α

= 1)10 0.00 0.00 0.00 (α

= 1)11 0.00 0.00 0.00 (α

= 0)12 18.63 21.24 18.63 (α

= 0)13 10.25 12.38 10.25 (α

= 0)14 32.54 24.32 24.09 (α

= 0.87)15 48.58 37.22 36.72 (α

= 0.84)16 28.91 25.42 25.42 (α

= 1)17 6.48 10.53 6.48 (α

= 0)18 17.00 21.66 17.00 (α

= 0)19 22.65 9.55 9.55 (α

= 1)

Collaborative Wins

Statistical Wins Hybrid Wins

Jha 21 May 2008 17

Collaborative Forecastworst among the three

Jha 21 May 2008 18

PID Collaborative RMSE Statistical RMSE Hybrid RMSE1 0.00 0.00 0.00 (α

= 0)2 2.31 2.04 1.82 (α

= 0.61)3 2.89 2.61 2.28 (α

= 0.58)4 19.69 3.83 3.83 (α

= 1)5 48.91 26.48 26.48 (α

= 1)6 4.62 4.71 4.56 (α

= 0.38)7 17.22 10.64 10.40 (α

= 0.86)8 1.15 1.31 1.01 (α

= 0.40)9 2.31 2.29 1.82 (α

= 0.51)10 25.36 6.55 6.55 (α

= 1)11 6.93 11.22 5.46 (α

= 0.30)12 30.31 5.57 5.57 (α

= 1)13 2.38 2.14 1.66 (α

= .56)14 5.29 7.06 5.25 (α

= 0.12)15 6.00 7.92 7.14 (α

= 0.84)16 20.91 15.65 15.65 (α

= 1)17 4.62 3.67 3.67 (α

= 1)18 285.12 23.12 23.12 (α

= 1)19 55.12 41.87 41.87 (α

= 1)20 5.16 4.02 2.91 (α

= 0.61)

Statistical Wins

Hybrid Wins

AgendaBackground

Product Hierarchy

Statistical Forecasting

Results

Conclusions

Jha 21 May 2008 19

ConclusionsThere is value to both statistical and collaborative forecasts.

A bottom-up and top-down approach in statistical forecasting helps to generate more accurate forecasts

In some scenarios application of simple statistical techniques yield better results than application of complex techniques.

Jha 21 May 2008 20

Jha 21 May 2008 21

Recommended