Beruflich Dokumente
Kultur Dokumente
216
Manufacturing firm:
A branch of a bank:
Considerations:
1. multiple inputs and outputs
2. Different measurement units (KPIs), non-financial metrics
3. (possibly) conflicting metrics and perspectives
217
Ways to rank/benchmark
• In some cases, a dominant metric exists
• Hrs/transaction (in call center) , $sales/sq.-ft. (mall and casino)
Problems:
In most organizations, many different mix of inputs and outputs
Some inputs/outputs cannot be valued in $ (non-profit)
Difference between efficiency and profitability.
218
Cost Profit
Store 1 (Lauren) $10K $30K
Store 2 (Emily) $10K $40K
219
220
221
222
223
224
30
B5
20 B2
B4
10
0
0 10 20 30 40
Loans
Which of the 5 branches can be considered efficient (i.e., can be ranked
best) with some scoring rules?
225
10
0
0 10 20 30 40
Loans
Can you say Bank 1 is better than Bank 5?
226
10
0
0 10 20 30 40
Loans
Are there any weights (on the output) that make B2 the best?
227
10
What would be the best scoring rule from the B2’s perspective?
If we use that particular scoring rule, what would be the score (relative
efficiency) for B2, if we use a score within [0,1]?
228
30
B5
20 B2
B4
10
0
0 10 20 30 40
Loans
(1) Which scoring rule will give B2 the best possible score that
ranges between 0 and 1?
229
30
B5
20 B2
B4
10
0
0 10 20 30 40
Loans
• DEA takes the view from one particular organization to find the best
scoring rule for that organization.
• DEA alone does not tell you what is the most fair and precise scoring rule
for benchmarking.
230
q No branch with same (or more) deposit vol. and more loan vol.
Efficient Frontier:
Deposits the line connecting efficient units
40
B1
B3
30 Organization is on efficient frontier
B2 B5
20 B4
10
Scores 1 (on a [0,1] scale) according to
0 some scoring rules.
0 10 20 30 40
Loans
231
20
B2
B4
B5 Branch 2: red line: 29.155
dotted line: 35.206,
10
0 Efficiency=
0 10 20 30 40
Loans
232
10
0
0 10 20 30 40
Loans
233
Deposits
40
B1
B3
30
HCU5=B5
B2 B5
20 B4
10
0
0 10 20 30 40
Loans
234
DEA Formulation
Graphical method cannot apply to cases with multiple inputs and outputs.
Is there a formal approach to measure the efficiency of DMU unit i?
Weighted sum of outputs
Efficiency of an organization =
Weighted sum of inputs
Multiple outputs j (yj): loan volume, deposit volume, profit
Multiple inputs k (xk): operating expense, labor
What is the best scoring rule (weights on inputs and outputs) for this
organization?
If we use that particular scoring rule, what is the relative efficiency of this
organization (on 0-1 scale)?
Decision variables
o j : weight given to output j
ik : weight given to input k
235
1. Gives the best score for B2 (Maximize the score for B2)
2. All organizations’ efficiencies are on [0,1] scale.
236
237
238
239
240
Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$G$16 Input value Efficiency 1 0.828125 1 1E+30 1
$G$17 Efficiency of B1 <= 1 Efficiency 0 0.15625 0 0.033333333 0.18
$G$18 Efficiency of B2 <= 1 Efficiency -0.171875 0 0 1E+30 0.171875
$G$19 Efficiency of B3 <= 1 Efficiency 0 0.671875 0 0.123287671 0.032258065
$G$20 Efficiency of B4 <= 1 Efficiency -0.209375 0 0 1E+30 0.209375
$G$21 Efficiency of B5 <= 1 Efficiency -0.28125 0 0 1E+30 0.28125
241
oL oD iE iF
Decisions 0.02 0.02 0.01 0 Branch 4
23 23 Efficiency 0.920
Constraints
Input value 0 0 100 6 1 = 1
Efficiency of B1 <= 1 10 31 100 5 -0.18000 <= 0
Efficiency of B2 <= 1 15 25 100 5 -0.20000 <= 0
Efficiency of B3 <= 1 20 30 100 5 0.00000 <= 0
Efficiency of B4 <= 1 23 23 100 6 -0.08000 <= 0
Efficiency of B5 <= 1 30 20 100 5 0.00000 <= 0
242
243
244
D. But, DEA does not tell you which ranking method is the most fair and/or
adequate to use for benchmarking.
245
Unit Profit (in $K) Satisfaction (10-point) Cleanliness (100-point) Labor Hrs (In FTE) Op. Costs (in $K)
1 5.98 7.70 92 4.74 6.75
2 7.18 9.70 99 6.38 7.42
3 4.97 9.30 98 5.04 6.35
4 5.32 7.70 87 3.61 6.34
5 3.39 7.80 94 3.45 4.43
6 4.95 7.90 88 5.25 6.31
7 2.89 8.60 90 2.36 3.23
8 6.40 9.10 100 7.09 8.69
9 6.01 7.30 89 6.49 7.28
10 6.94 8.80 89 7.36 9.07
11 5.86 8.20 93 5.46 6.69
12 8.35 9.60 97 6.58 8.75
246
247
248
249
Low High
Eff. Eff.
Low Profit
251
Limitations
252
253
NON-LINEAR OPTIMIZATION
Unconstrained optimization
Constrained optimization
Quadratic programming
Applications
254
Non-linear problems
What if the functional relationship is non-linear?
Non-proportional relationships between decisions and the
metric.
255
Non-proportional Relationships
Diminishing rate of returns
x , ln x, piece linear, 1 - e-X/RT etc.
e.g., quantity discount,
risk aversion
Others.
quadratic, polynomial , trigonomet ric.
256
257
Convexity
• What is a convex set?
258
Example: x2
• A continuously differentiable function of one variable is convex
on an interval if and only if the function lies above all of its
tangents: f(x)>=f(y)+f’(y)(x-y)
259
Function Convex/Concave/Neit
her
|x|
ex
1/x
ax+b
260
Concave functions
• A function f is called concave if –f is convex.
• Geometry: For a convex function f lies below the line joining
f(x) and f(y) and for a concave function f lies above the line.
261
262
264
Problem Statement
• Generic formulation
265
Example 1: EOQ
A retailer is planning her yearly inventory strategy for a
commodity which sells a steady rate thorough the year and for
which she estimates a yearly demand of D units. The storage
cost is $S per unit and the cost of ordering is $C per order.
Assuming that there is no lead time, how many orders and of
which size must be placed to reduce the total
inventory costs ?
266
Example 1: Model
• Suppose the retailer places an order size x, then must place D/x
orders at a cost C * (D/x)
• The average inventory is x/2 therefore the holding costs are
S * x/2
• Optimization problem
In reality, must have the constraint x >=0 but ignored it since one
expects the solution to be in [0, ]
267
EOQ Model
• Check convexity
• Find minimizer
268
269
• The goal is to find the parameter values for the model that
"best" fits the data. The least squares method finds its
optimum when the sum of squared residual is a minimum.
270
Example 2: Model
• The squared error at period t is given by
For t=4,
Optimization problem
minimizea
271
Excel Model
272
Constrained Optimization
• We now have constraints on the decision variables
273
274
275
278
279
Decision variables?
Covariance
Macys INTC
Macys 19.359
INTC 9.697 14.883
280
R = 0.4XM + 0.6XINT
Macys INTC
(i)E[R] =
Avg 0.911 0.556
(ii) Var[R] =
Covariance
Macys INTC
Macys 19.359
INTC 9.697 14.883
281
Covariance
Optimizing the portfolio Macys INTC
Macys 19.359
Minimizing the variance of the portfolio return INTC 9.697 14.883
Macys INTC
Constraints:
Avg 0.911 0.556
282
Var= , Stdev =
283
Suppose you allocate 50% to Macys, 30% to Intel, and 20% to PB.
284
Suppose you allocate 50% to Macys, 30% to Intel, and 20% to PB.
285
286
Suppose you allocate 50% to Macys, 30% to Intel, and 20% to PB.
287
Non-linear programming
Portfolio Selection
This is the essence of portfolio optimization aka Markowitz
portfolio.
288
Price
200 500 1000
289
Price Differentiation
• Suppose now:
• We charge two different prices: $300 and $750
• Anybody who is willing to pay $750 or more purchase at $750.
• Anybody who is willing to pay more than $300, but less than 750
purchase at $300.
• The sellerʼs profit will now be as follows:
Demand
800 Profit
700 = (300 – 200) × 450 + (750 – 200) × 250
These customers = $182,500
pay $300 each.
These 250
customers pay
$750 each. Price
200 300 750 1000
290
Price Differentiation
Demand
800
700
500
250
Price
200 300 500 750 1000
291
• What if customers who are willing to pay $750 find a way to purchase at
the lower price of $300 (cannibalization)?
• Arbitrage:
• What if there existed a group of speculators who decided to buy at $300
and sell to others at $750?
292
293
303
304
305
Optimal price =
# GA tickets sold =
Member attendance=
General public attendance =
Revenue =
306
307
308
309
310
311
312
313
314
Optimal # seats =
Optimal member price =
Optimal standing-public price
Optimal seated-public price =
# seated tickets sold =
# standing tickets sold to members=
# standing tickets sold to general public=
Revenue =
315
316
Segmentation
• When implementing differentiated prices, one of the most
important design decisions is determining the customer
segmentation.