Scribd
Upload
7 views2 pages

Factory Response to Widget Price Change

The factory produces widgets of one colour per day based on orders received. It must decide the colour in the morning to fulfill evening orders. Red has 100 units in stock and average daily orders of 50. Green has 150 units and 100 average orders. Blue has 50 units and 10 average orders. The first order is for 40 blue widgets. Choosing green minimizes expected unfilled orders at 10.6, compared to 22.7 for red and 29.4 for blue. The document then discusses fluctuations in orders and stable distributions to model them.

Uploaded by

JH
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views2 pages

Factory Response to Widget Price Change

The factory produces widgets of one colour per day based on orders received. It must decide the colour in the morning to fulfill evening orders. Red has 100 units in stock and average daily orders of 50. Green has 150 units and 100 average orders. Blue has 50 units and 10 average orders. The first order is for 40 blue widgets. Choosing green minimizes expected unfilled orders at 10.6, compared to 22.7 for red and 29.4 for blue. The document then discusses fluctuations in orders and stable distributions to model them.

Uploaded by

JH
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

The Widget Problem

Setup: factory produces one colour of widgets every day, orders for different colours come during the day,
want to fulfill them every evening. Task: decide which colour in the morning.
daily production: D = 200.
R G B decision?
(A) in stock: S1 = 100 S2 = 150 S3 = 50
(B) avg daily order total: A1 = 50 A2 = 100 A3 = 10
(C) avg daily number of orders: N1 = 2/3 N2 = 10 N3 = 1/2
(D) first order: 40

(A): no knowledge about orders, so “best guess” is equal probability: B would run out first

(B): state: daily total volume: (n1 , n2 , n3 ), ni = 0, 1, . . . . Know: hni i = Ai .


∞ ∞
! ∞
! ∞
!
X X X X
Z(λ1 , λ2 , λ3 ) = e−λ1 n1 −λ2 n2 −λ3 n3 = e−λ1 n1 e−λ2 n2 e−λ3 n3
n1 ,n2 ,n3 =0 n1 =0 n2 =0 n3 =0
| {z }
1
1−e−λ1

1 −λ1 n1 −λ2 n2 −λ3 n3


p(n1 , n2 , n3 ) = e = p1 (n1 )p2 (n2 )p3 (n3 )
Z
∂ ln Z 1
hni i = − = λi
∂λi e −1
1
Zi (λi ) = = Ai + 1
1 − e−λi
 ni
1 Ai
pi (ni ) =
Ai + 1 Ai + 1
Loss function: volume of unfilled orders
If we have chosen R:

L1 (n1 , n2 , n3 ) = R(n1 − S1 − D) + R(n2 − S2 ) + R(n3 − S3 )


where ramp function:

x if x ≥ 0
R(x) =
0 if x ≤ 0
similarly

L2 (n1 , n2 , n3 ) = R(n1 − S1 ) + R(n2 − S2 − D) + R(n3 − S3 )


L3 (n1 , n2 , n3 ) = R(n1 − S1 ) + R(n2 − S2 ) + R(n3 − S3 − D)

Expectation of loss:

10
∞
X
hL1 i = L1 (n1 , n2 , n3 )p(n1 , n2 , n3 )
n1 ,n2 ,n3 =0
X∞ ∞
X ∞
X  
= L11 (n1 )p1 (n1 ) · p2 (n2 ) · p3 (n3 ) + . . . L12 . . . + . . . L13 . . .
n1 =0 n2 =0 n3 =0
| {z } | {z }
1 1
 S1 +D+1  S 2  S 3
1 2 A1 A2 A3
= (A1 + 1) +A2 + A3
A1 + 1 A1 + 1 A2 + 1 A3 + 1
| {z }
S
1 +D

A1
A1 A1 +1

where we used
∞
X xn+1
(i − n)xi =
i=n
(x − 1)2
plugging in values:

hL1 i ≈ 0.131 + 22.48 + 0.085 ≈ 22.70


hL2 i ≈ 6.9 + 3.07 + 0.085 ≈ 10.6
hL3 i ≈ 6.9 + 22.48 + 4 · 10−10 ≈ 29.38

Fluctuations
Central limit theorem: suppose Xi are iid (independent identically distributed) random variables, with
hXi i = µ and Var(Xi ) = σ 2 . Then
Sn
z }| {
def X1 + · · · + Xn −nµ D
Zn = √ → N (0, 1)
nσ

where convergence in distribution:

lim P (Zn < z) = P (ζ < z) where ζ is std normal


n→∞

Cauchy distribution:
1 1
f (x) = or f (x) =  2 
π(1 + x2 ) 
x−x0
πγ 1 + γ

Stable distributions:

X1 ∼ Fam(Θ1 ), X2 ∼ Fam(Θ2 ) ⇒ aX1 + bX2 ∼ Fam(Θ3 ) + c

then Fam is stable


Levy: r
c e−c/(2x)
f (x) =
2π x3/2
Levy flight increments: f (x) ∼ 1/|x|α+1 , where 0 < α < 2.

11

You might also like