Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory...
Transcript of Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory...
![Page 1: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/1.jpg)
Independent Demand Models
• Non Linear (Chemical Industry - take or pay)• Deterministic Simulation (make to stock - lumpy
demand)• Mathematical Programming (family structure -
near optimum)• Heuristic (make to stock - sequence independent)• Heuristic (Make to stock - sequence dependent)
![Page 2: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/2.jpg)
Additional Models:
Capacitated MRP (finite planning for dependent demand)
Continuous Replenishment Systems (optimal truck loading)
![Page 3: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/3.jpg)
Welch’s has a large amount offorecast error and bias
As a result, Welch’s has experienced a breakdownin finite capacity planning
Often SKEP shows we are out of capacity (red) in the locked period and beyond, when the real causeis a forecast with high variability or bias
![Page 4: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/4.jpg)
Typical Forecast Errors at Welch’s,
• Are often quite dramatic• Seldom follow a normal distribution, even over
extended periods of time• Show extreme bias, with limited patterns of
adaptation through time
• Production planners deal with weekly variability• We measure error by dividing the monthly
forecast by the number of weeks in a month and comparing to actual shipments for the week
![Page 5: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/5.jpg)
A Detailed Look at Welch’s Forecast Error Data
We calculate the forecast indicators and list:
a. Forecast error by product by manufacturing plant(three tables)
b. Forecast error for a single product manufactured at all plants in the Welch network(five tables)
c. No table lists forecast errors by manufacturing lined. Some of Welch’s forecast error results from
improper plant splits rather than high levelforecast error (product manager level)
![Page 6: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/6.jpg)
Negative Bias in the Forecast
• FETS varies between 0 and 1• Consistent oversell of the forecast• Implication: out of stock before next
scheduled production run, forced break in production sequence to maintain customer service
• A deadly situation in cases where capacity utilization is high
![Page 7: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/7.jpg)
Example of Negative Bias (Oversell) WPD21100 at KW, FETS = -.65
0200040006000800010000120001400016000
9/1/97
10/1/97
11/1/97
12/1/97
1/1/98
2/1/98
3/1/98
4/1/98
5/1/98
6/1/98
7/1/98
8/1/98
9/1/98
10/1/98
11/1/98
weeks
case
s sh
ippe
d pe
r w
eek
![Page 8: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/8.jpg)
Positive Bias in the Forecast
• FETS varies between 0 -1 • Forecast is frequently higher than actual• Implication - by planning production to the
forecast, we have too much inventory• This situation is deadly when there are
warehouse storage limits on products
![Page 9: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/9.jpg)
Example of Positive Bias (undersell) WPD19200 at LT, FETS = .83
010002000300040005000600070008000
9/1/97
10/1/97
11/1/97
12/1/97
1/1/98
2/1/98
3/1/98
4/1/98
5/1/98
6/1/98
7/1/98
8/1/98
9/1/98
10/1/98
11/1/98
weeks
case
s sh
ippe
d pe
r w
eek
![Page 10: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/10.jpg)
High Variability Between Forecast and Actual
• Demand in relation to the forecast means almost nothing
• Implication - the need for high safety stocks• It is hard to put a limit on how much safety stock
is needed• Creates substantial uncertainty when several
products on a production line (with finite capacity) experience high variability
• More inventory required!!!!
![Page 11: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/11.jpg)
Example of High Variability WPD22900 at KW, TICF = .94
010002000300040005000600070008000900010000
9/1/97
10/1/97
11/1/97
12/1/97
1/1/98
2/1/98
3/1/98
4/1/98
5/1/98
6/1/98
7/1/98
8/1/98
9/1/98
10/1/98
11/1/98
weeks
case
s sh
ippe
d pe
r w
eek
![Page 12: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/12.jpg)
Low Variability Between Forecast and Actual
• A stable situation• Finite capacity calculations are believable• Occasional statistical “blip” seldom, if ever,
is maintained through to the next period • “Blips” are sometimes predictable (end of
the year push)• LOW INVENTORY REQUIRED
![Page 13: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/13.jpg)
Example of Low Variability WPD21100 at NE, TICF = .3
0100002000030000400005000060000700008000090000100000
9/1/97
10/1/97
11/1/97
12/1/97
1/1/98
2/1/98
3/1/98
4/1/98
5/1/98
6/1/98
7/1/98
8/1/98
9/1/98
10/1/98
11/1/98
weeks
case
s sh
ippe
d pe
r w
eek
![Page 14: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/14.jpg)
Cost of Forecast Error
• Less forecast error means less safety stock• Sample 4-5 major products produced at all
manufacturing plants• Compare safety stock (SS) calculation with
current forecast error to SS calculation with min error (30% absolute error)
![Page 15: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/15.jpg)
Conclusion• Total safety stock is 47,074 cases for current
forecast error • Total safety stock is 35,137 cases for “ideal”
forecast error• Ideal forecast error results in 1/3 less packed
product (PP) inventory• Assume Average PP inventory = $24 MM• A 1/3 reduction = $8 MM @ 10% carrying cost =
$800,000
![Page 16: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/16.jpg)
Conclusion (continued)
• With less need for safety stock, supplies and ingredients inventory also decreases:
estimated decrease (conservative) = $0.2 MM
• Less safety stock also mean fewer “rapid sales” situations and less “obsolete” material in inventory
estimated decrease (conservative) = $1.0 MM
• First cut at total decrease in cost:Savings of about $2.0 MM
![Page 17: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/17.jpg)
R.P. Without Safety Stock
(u)(t)
R.P.P
Inv.Level
t Time
![Page 18: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/18.jpg)
R.P. With Safety Stock
Time
Inv.Level
P
SS
![Page 19: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/19.jpg)
Safety Stock in Finite Planning Systems
• User input, “days of supply,” no direct link to customer service levels
• Calculation of safety stock based on forecast• A “lumpy” forecast produces a dynamic safety stock
through time• Production planners have a hard time determining
safety stock levels for the 200-300 sku’s• Improper safety stock levels negatively impacts a)
production timing, b) production lot size and c) production sequencing
![Page 20: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/20.jpg)
Archetypal Finite Planning Systems,• Do not integrate statistical safety stock planning into
algorithms or heuristics• Make no provision for type 1 or type 2 customer
service levels (ability to meet demand Vs total percentage of cases shipped)
• Do not account for forecast bias• Assume independent demand is “deterministic”• Accept forecast at face value• Can not find optimal solutions for sequencing and lot
sizing problems under situations with dynamic safety stock
![Page 21: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/21.jpg)
Safety Stock is Defined as:Demand through lead time depending on:
1. Length of lead-time2. Forecast error3. Forecast bias4. Level of customer service
The nature of forecast error has a great impact on safety stock
Each forecast has its own “fingerprint,” not every forecast is bad.
The only practical way to deal with safety stock andforecast error in finite capacity planning isthrough “layering” of models.
![Page 22: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/22.jpg)
TICF is a measure of variability**larger means greater forecast error, positive and negative**similar to a standard deviation
FETS is a measure of forecast bias**0 to 1 the forecast is high compared to actual**0 to -1 the forecast is low compared to actual**0 the forecast is neutral, errors randomly distributed
S is a coefficient applied to safety stock**FETS between 0 and 1 gives S is less than 1**no adjustment for negative bias, we assume the
forecast will be updated
![Page 23: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/23.jpg)
The Calculation of Statistical Safety Stock
SS = (S ) x (k ) x (T IC F) x (u ) x (t)
W here:
u = fo recast dem and per day
t = lead tim e
k = serv ice level m ultip lier
s = suppression facto r (straigh t line) = 1 – FE T S
TIC F u i x i u i ni
n= −
=∑ ( ) ( ) / ( ) / /1
u (i) = past w eekly dem andsx(i) = past actual w eekly salesn = num ber o f periods
F E TS u i x i u i T IC Fi
n= −
=∑ ( ( ) ( ) / ( ) / /1
If the forecast improves, the model calculates less safety stock.Do we need to smooth the bias in SS calculations?
![Page 24: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/24.jpg)
The Formula for Statistical Safety Stocks With Bias Adjustment*
SS = (u)(t) x TICF x (k) x (s)
*Krupp, J.A.G. “Effective Safety Stock Planning,” ProductionInventory Management Journal, Third Quarter, 1982.
![Page 25: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/25.jpg)
Trigg Indicator
• Used to identify a “strange” forecast• Spread of forecast errors is non-random• Smoothing factor of .1 (β in the equations)
used for calculation of the indicator (significant weight given to past observations)
• A calculated value of T greater than .51 shows a forecast with peculiar bias
![Page 26: Independent Demand Models - MITweb.mit.edu/edmund_w/www/Managing Production and Inventory (7).pdf · Capacitated MRP (finite planning for ... • Calculation of safety stock based](https://reader031.fdocuments.us/reader031/viewer/2022021817/5a9b1ba17f8b9adb5c8dd3c3/html5/thumbnails/26.jpg)
The Formula for a Tracking SignalDeveloped by Trigg*
E t e t E t
M t e t M t
T t E t M t
( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) / ( )
= + − −
= + − −
=
β β
β β
1 1
1 1
If forecasts are unbiased, the smoothed error E(t) should be smallcompared to the smoothed absolute error M(t).
*Trigg, D.W. “Monitoring a Forecasting System.” Operational ResearchQuarterly 15, 1964, pp. 271-74.