Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009...

24
Schedule On Thursdays we will be here in SOS180 for: 19-02-2009 (today) – 26-02-2009 – 05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Transcript of Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009...

Page 1: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Schedule

• On Thursdays we will be here in SOS180 for:– 19-02-2009 (today)– 26-02-2009– 05-03-2009

• Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Page 2: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Today

• We will write and solve the corresponding LP to analyze a simple DEA problem.

• Here is our data set:

DMU A B C D E F G

Input I1 4 7 8 4 2 10 3

I2 3 3 1 2 4 1 7

Output O 1 1 1 1 1 1 1

Page 3: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Graphically...

Page 4: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Mathematical modelling

N

iiki

M

jjkj

k

k

j

i

ik

jk

Iv

Ou

E

E

ju

iv

kiI

kjO

MjM

NiN

KkK

1

1

1) and 0(between DMU of efficiency

output on weight

input on t weigh

DMU from input of level observed 0

DMU from output of level observed 0

,...,1 outputs#

,...,1 inputs #

,...,1 (DMUs) units operating# Data

Model variables

Page 5: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Niv

Mju

KkIvOu

Iv

ts

OuE

i

j

N

iiki

M

jjkj

N

iiei

je

M

jje

,...,1 ,0

,...,1 ,0

,...,1 ,

1

..

max

11

1

1

Linear Program (LP):

Page 6: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Obtain file(s):

• Copy and paste the excel file to your own accounts:F:\COURSES\UGRADS\INDR\INDR471\SHARE\Labs\ lab1-LP-

raw.xls

• Also open the power point file:F:\COURSES\UGRADS\INDR\INDR471\SHARE\labs\DEA-Lab-18-

02-09.ppt

Page 7: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

1

Solve the corresponding LP for DMUs A in Excel. Keep the sensitivity report.

Niv

Mju

KkIvOu

Iv

ts

OuE

i

j

N

iiki

M

jjkj

N

iiei

je

M

jje

,...,1 ,0

,...,1 ,0

,...,1 ,

1

..

max

11

1

1

Page 8: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

SUMPRODUCT

Multiplies corresponding components in the given arrays,

and returns the sum of those products.

Syntax

SUMPRODUCT(array1,array2,array3, ...)

Array1, array2, array3, ...   are 2 to 30 arrays whose

components you want to multiply and then add.

Remarks•The array arguments must have the same dimensions. •If they do not, SUMPRODUCT returns the #VALUE! error value. •SUMPRODUCT treats array entries that are not numeric

as if they were zeros.

Page 9: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Excel solver

• Tools Add-in Excel solver

• Options International Decimal and Thousands (for later use, but…)

Page 10: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

1.a

What is the efficiency of DMU A? What are the optimal input-output weights for DMU A? Does the solution show whether DMU A has purely technical (or ratio) and/or mix inefficiencies? How? Why? Explain.

Page 11: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

1.b

What is the level of input for DMU A to be ratio (or purely technical) efficient?

Note its meaning: If the DMU becomes ratio efficient, it can produce the same level of output with a decreased amount of input, while keeping the input ratios the same.

iee IE

inputs for A* efficiency of A:4*0.8517 = 3.4286 (input 4’tu 3.42 olmalı)3*0.8517 = 2.5714

Page 12: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

1.c

1. Which constraints are binding? 2. What are the shadow prices of the constraints? 3. Use this information to determine the

reference set for each of the DMUs. 4. Can we use the above information to find

target input/output levels for DMU A?

Note: A binding constraint have a slack value of 0. The shadow prices of non-binding constraints are always 0 (Why?). The values of slacks for the constraints are given in “Final Value” of “Constraints” section in the

sensitivity report.

Page 13: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Definition: Reference Set

For an inefficient DMUe, we define its reference set

Re,

based on the max-slack solution as obtained in phases

one and two, by (binding olanlar reference set oluyor.)

Re= {kє{1, …, K} | k*>0}

e

N

iiki

M

jjkj RkIvOu

,11

Dual variable vs its constraint in LP

Page 14: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Hypothetical composite unit (HCU) and target setting for inefficient for

DMU e

e. DMUfor set reference the: where

M,1,...,j ˆ

N,1,...,i ˆ

*

*

e

jkRk

kje

ikRk

kieeie

R

OλO

IλIEI

e

e

Input targets:

Output targets:

CCR projections.

Page 15: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

For DMU A to be efficient:

(Lamda=shadow price)0.7143*inputs for unit D + 0.2857 * inputs for unit E

=0.7143* (4) + 0.2857 * (2) (2) + (4)

= 3.42862.5714

current inputs: 4, 3

inputs for A* efficiency of A:4*0.8517 = 3.42863*0.8517 = 2.5714

Page 16: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Meaning of shadow prices in DEA

Niv

Mju

KkIvOu

Iv

ts

OuE

i

j

N

iiki

M

jjkj

N

iiei

je

M

jje

,...,1 ,0

,...,1 ,0

,...,1 ,0

1

..

max

11

1

1

LPe

Niv

Mju

IvOu

fkkIvOu

Iv

ts

OuE

i

j

N

iifi

M

jjfj

N

iiki

M

jjkj

N

iiei

fje

M

jje

,...,1 ,0

,...,1 ,0

,

, ,0

1

..

max

11

11

1

1

LPe with an increase in the RHS

Page 17: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

1.d

Do you observe binding constraint(s) with a shadow price of 0 for any of these DMUs? If yes, which one? Is it efficient or inefficient?

Binding constraints with a shadow price of 0 indicate the existence of multiple optima.

Page 18: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

2.a

What is the efficiency of DMU G? What are the optimal input-output weights for DMU G? Does the solution show whether DMU C has purely technical (or ratio) and/or mix inefficiencies? How? Why? Explain.

Page 19: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

2.b

What is the level of input for DMU G to be ratio (or purely technical) efficient?

Note its meaning: If the DMU becomes ratio efficient, it can produce the same level of output with a decreased amount of input, while keeping the input ratios the same.

iee IE

inputs for G* efficiency of G:3*0.6667 = 27*0.6667 = 4.6667

Page 20: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

2.c

1. Which constraints are binding? 2. What are the shadow prices of the constraints? 3. Use this information to determine the

reference set for each of the DMUs. 4. Can we use the above information to find

target input/output levels for DMU G?

Note: A binding constraint have a slack value of 0. The shadow prices of non-binding constraints are always 0 (Why?). The values of slacks for the constraints are given in “Final Value” of “Constraints” section in the

sensitivity report.

Page 21: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

For DMU G to be efficient:

1 * inputs for unit E

(2)= (4)

current inputs: 3, 7

inputs for G* efficiency of G:3*0.6667 = 27*0.6667 = 4.6667

Page 22: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

2.d

Do you observe binding constraint(s) with a shadow price of 0 for any of these DMUs? If yes, which one? Is it efficient or inefficient?

Binding constraints with a shadow price of 0 indicate the existence of multiple optima.

Page 23: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Our goal

• To compute the efficiency of each DMU (identify both ratio and mix inefficiency, if present)

• To set target levels for each DMU, which will make them efficient

• To identify efficient DMUs which have similar input-output levels to an inefficient DMU e, so that DMU e can learn from these specific efficient DMUs.

Does the LP formulation fulfill these goals?

Page 24: Schedule On Thursdays we will be here in SOS180 for: –19-02-2009 (today) –26-02-2009 –05-03-2009 Homework 1 is on the web, due to next Friday (17:00 27-02-2009).

Niv

Mju

KkIvOu

Iv

ts

OuE

i

j

N

iiki

M

jjkj

N

iiei

je

M

jje

,...,1 ,0

,...,1 ,0

,...,1 ,

1

..

max

11

1

1

LPe for CCR and its dual DLPe:

edunrestrict:

,...,1 ,0

,...,1 ,

,...,1

..

min

1

1

Kk

NiII

MjOO

ts

k

ie

K

kikk

K

kjejkk

LPe DLPe

Linear combinationof observedinput values.

Linear combinationof observedoutput values.

Output of DMU e

Effectiveinput of DMU e