Consulting
• Third-party data
– Industry research (A. C. Nielsen, etc.)
– Macro-economic data
• Survey/experimental data
– Questionnaires
– Live experiments
• Expert Judgment
• Price of alternative i
• Deterministic choice: customer picks the alternative with the highest Net
Utility (surplus)
• Histogram
Max WTP
40
35
30
25
20
15
10
0
25 50 75 100 125 150
• Nonparametric estimation
– Use raw data to directly
• Empirical histogram of demand volume
• Empirical histogram of reservation prices
– Pros: no assumptions, directly from observed data
– Cons: cannot extrapolate beyond data, hard to optimize, over-fitting
• Parametric estimation
– Assume a demand model with modest number of parameters, and estimate
parameters from data
• Linear: D(p) = a – bp
• Normal : D ~ N(µ, σ2)
– Pros: concise description, mathematically tractable, can extrapolate beyond data
– Cons: assumptions on the form of response may not be valid
• Demand Functions D
– Non-negative
– Downward sloping
– Continuous (hopefully)
– Differentiable (ideally)
• Exceptions not considered
P
– Luxury goods
– Giffen goods
– Goods that its price affects the perceived quality
• Measures of price effect
– Price sensitivity: increase p by $1, demand decreases by ?
• Unit dependent
– Price elasticity: increase p by 1%, demand decreases by ?%
WTP
Distribution
Uniform Exponential
D = a – bp D = exp(a-bp) D = a p-e
|ϵ|
|ϵ|
|ϵ|
• Examples:
– Multiple versions of a product
– Competing products
• Models:
– Multi-product demand functions
– Discrete choice models
– Parameter estimation can be done by regressing each demand against both prices
utility for i = ui – pi + ξi
𝑉. 67 897
Pr(chooses i) = , 𝑉. = 𝑒 :
1 + 𝑉1 + ⋯ + 𝑉3
Price P $50
0.9
0.8
0.7
0.6 deterministic
s=1
0.5
s=10
0.4 s=100
s=10000
0.3
0.2
0.1
0
0
4
8
12
16
20
24
28
32
36
40
44
48
52
56
60
64
68
72
76
80
84
88
92
96
100
104
108
112
116
120
124
128
132
136
140
144
148
Aug 2016 DSC4213 Session 4 - Prof. WANG Tong 21
NYHC B: Estimate MNL from WTP data
ui = average(WTPi)
• Prediction
– Calculate Vi = exp( (ui – pi) / s)
– Purchase probability
• Model
– Independence of Irrelevant Alternatives (IIA)
• Optimization
– Not trivial
• Extentions
– Probit model: multivariate Normal ξi
– Generalized extreme value (GEV) model: allow correlation in ξi
– Nested logit: allow hierarchical choices
– Mixed logit: individual variations
• NYHC Data
– Price =$50 for all slots
– Observed choices
Noon-
Client 6am-9am 9am-noon 2pm 2pm-5pm 5pm-9pm 9pm-close No-purchase
1 1 0 0 0 0 0 0
2 0 0 0 0 1 0 0
3 1 0 0 0 0 0 0
4 0 0 0 0 1 0 0
5 0 0 0 0 1 0 0
6 0 0 1 0 0 0 0
7 0 0 0 0 0 1 0
8 0 0 0 0 0 1 0
9 0 0 0 0 1 0 0
10 0 0 0 0 0 1 0
11 0 0 0 0 1 0 0
12 0 0 0 0 0 0 1
13 0 0 0 0 1 0 0
14 0 0 0 0 0 0 1
15 0 0 0 1 0 0 0
• Quick example
– Two possible models: f(x | u1) and f(x | u2)
– Data: x1, x2, …, xn
– Likelihood of the two models:
• f(data|u1) = f(x1|u1) f(x2|u1) … f(xn|u1) = L(u1|data)
• f(data|u2) = f(x1|u2) f(x2|u2) … f(xn|u2) = L(u2|data)
– Choose the one with max likelihood
• Solving max L(𝑢, 𝑠|data) or max log L 𝑢, 𝑠 data yields the maximum likelihood
6,: 6,:
estimate
s= 9.888807
9am- Noon- 5pm- No
6am-9am noon 2pm 2pm-5pm 9pm 9pm-close Purch.
Price $ 50.00 $ 50.00 $ 50.00 $ 50.00 $ 50.00 $ 50.00
Total Log
uj $ 55.05 $ 34.08 $ 44.95 $ 39.14 $ 55.81 $ 50.64 Likelihood
Net util. $ 5.05 $ (15.92) $ (5.05) $ (10.86) $ 5.81 $ 0.64 0 -75.98299317
vj 1.666646 0.199999 0.599996 0.333329 1.79998 1.066657 1
pj 0.249999 0.03 0.09 0.05 0.269999 0.16 0.150001
• Aggregate demand
Demand 25 3 9 5 27 16 15
• Likelihood is
H