Date post: | 15-Mar-2018 |
Category: |
Documents |
Upload: | hoanghuong |
View: | 218 times |
Download: | 4 times |
INSEADIoana Popescu
DPRM3
DYNAMIC PRICING & REVENUE MANAGEMENT
Professor Ioana Popescu
Session 3: Demand estimation. Static pricing with constraints.
• Demand estimation: transaction data & survey methods• Static price optimization with budget constraints. • Consumer perception of price
© I.Popescu 2010 DPRM 3
DPRM Overview
Basic Frameworks and tools Industry To do
1. Introduction to revenue management
2. Multi‐pricing in segmented markets Entertainment Poll 1
Demand Estimation & Pricing
3. Static pricing with constraints Adwords/Congestion Group 1
4. Dynamic pricing: markdowns Retail/Consumer Goods Strategy
5. Benefit assessment Fashion
Revenue Management
6. Managing risks Airlines/Hotels Poll 2
7. Multi‐resource management Media/Clouds Poll 3
8. RM for price takers Leasing/Transport Group 2
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
This week
The DPRM Process
4. Manage Price
Decisions
5. ManageInventory/Availability
2. Design product line 3. Forecast
demand
1. Segmentthe market
Today
our focus
© I.Popescu 2010 DPRM 3
Some background
• Demand modeling, estimation and forecasting is one of the most important implementation challenges in DPRM practice. Major development and operational time and effort are spent on demand forecasting and estimation
• Industry studies suggest a 20% reduction in forecast error can translate into a 1% incremental increase in revenue (Poelt AGIFORS 1998)
• “Forecasting” often connotes a single‐number, but in DPRM we need entire demand functions, plus information on the degree of demand uncertainty (demand distributions) to make good decisions
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Demand modeling & forecasting within RM
Forecasting moduleBookingsNo-shows
CancellationsRatings (media)Fares (airlines)
Groups utilizationPrice sensitivity
Transactiondata
Optimization moduleAllocationsBid Prices
Overbooking LimitsMarkdownsPromotions
User interface
Marketdata
Productsdata
Pricingdata
Reservation/salessystem
Manager Analyst Analyst
Un
con
stra
inin
g
Methods for estimating the price‐response function
Sales Data 3rd party data
Surveys/ Experiments
Experts judgements
POS data, loyaltlyprogram data, clickstream data
Panels (ACNielsen)
MIDS (mkt info data tapes – airlines)
Questionnaires, Live experiments
Conjoint
Company experts, Salesforce
Price response estimation methods
INSEADIoana Popescu
DPRM3
Source : Simon-Kucher & Partners
Classification of estimation methods
Internal External
Historical data
Company experts
Sales force
Individual customers
Many customers (>30 per segment)
In-dept interviews 0 + + +
Expert judgment workshops
+ + +
Focus Groups 0
Structured questionnaire
+ + +
Statistical analysis
+ + 0
Choice modeling + +
In-market tests 0 + +
Met
hods
Sources
Dire
ctio
nal
Qua
ntita
tive
+ + = Great …+ = Medium …0 = Limited …approach to gain
insights for identifying profit
opportunity
EDM case – managing an advertising budget
Half the money I spend on advertising is wasted. The trouble is I don’t know which half.
John Wanamaker, retail store owner
INSEADIoana Popescu
DPRM3
Search advertising
Traffic Estimator Data (see EDM.xls file)
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Estimation based on transaction data: EDM
Max CPCb
CPC C(b)CPD X(b)
Profit
Σ X(b)*conv.rate* Conv. profit before advertising ‐ X(b) *C(b)Maximize:
Estimate:
Constraint: Total Cost = Σ X(b)*C(b) ≤ Budget
Decide (for each keyword):
© Prof. Ioana Popescu
Step 1: EDM Data analysis
• Use Google Traffic Estimator data to estimate cost (CPC) & demand (CPD) based on max_CPC bid for each keyword
Wheelchair Rental CPC estimation
y = 0.8424x + 0.011 R2 = 0.999
$-
$0.20
$0.40
$0.60
$0.80
$1.00
$- $0.20 $0.40 $0.60 $0.80 $1.00 $1.20
max CPC
estim
ated
CPC
Wheelchair Rental CPD estimation
y = 14.794x + 3.3162 R2 = 0.9693
0
5
10
15
20
$- $0.50 $1.00 $1.50
max CPC
estim
ated
CPD
INSEADIoana Popescu
DPRM3
© Prof. Ioana Popescu
Step 2: EDM constrained optimization model
• Data : – volume X(b) & cost C(b) from traffic estimator data analysis
– conversion rates from historical data
– budget
• Objective:
maximize profit = Σi X(bi)*conv.ratei* $100 ‐ X(bi) *C(bi)
• Decision variables: b1, b2, b3 [max_CPC for each keyword i ]
• Constraints: Σi X(bi)*C(bi) ≤ $10 [Total Cost ≤ Budget]
© Prof. Ioana Popescu
EDM summary of historical data
Term Clicks Total Cost CPC Conversions Conversion % Cost Per Conversionrent wheelchair 1024 826.35 0.81$ 11 1.07% 75.12$ rent wheelchairs 2880 2761.94 0.96$ 36 1.25% 76.72$ rental wheelchair 816 730.73 0.90$ 9 1.10% 81.19$ rental wheelchairs 953 946.66 0.99$ 9 0.94% 105.18$ wheelchair rental 5311 4190.64 0.79$ 78 1.47% 53.73$
10984 9456.32 0.86$ 143 1.30% 66.13$
INSEADIoana Popescu
DPRM3
EDM: Multi‐Pricing with Constraints
STEP 1: Forecast demand (CPD) and cost (CPC) based on max_CPC bid for each keyword– build parametric models (data analysis → regression)– non‐parametric models works w/o constraints
STEP 2: Using the forecasted models, set up an optimization model (Solver) to determine bids for each keyword to optimize total profits subject to budget constraints
© I.Popescu 2010 DPRM 3
Estimation based on wtp data
• Vertigo
• Kilimanjaro
• London Congestion Charge
INSEADIoana Popescu
DPRM3
Pricing a Kilimanjaro hike
• Data analysis: estimate a demand model based on your data
• Optimization: find the price that maximizes revenue/profit
© I.Popescu 2010 DPRM 3
Parametric estimate based on linear demand (Kili P5)
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Parametric estimate based on linear demand (Kili P5)
Linear demand: d(p)=a‐bpMaximize: Revenue = p (30.55 – 0.0114 p) ⇒ p*= 1334
(vs. 900 empirical)
© I.Popescu 2010 DPRM 3
Parametric estimate: log‐linear demand (Kili P3)
Log‐linear demand: d(p)= exp(a‐bp)Maximize: Revenue rate= p * exp(– 0.001p) ⇒ p*=$ 970
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Parametric estimate based on linear demand (Kili P3)
Linear demand: d(p)=a‐bpMaximize: Revenue rate= p (0.8896 – 0.0005 p) ⇒ p*= 873
(vs. 500 empirical)
W illingness to pa y dis tribution
y = -0.0005x + 0.8869R2 = 0.7042
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
0 500 1000 1500 2000 2500 3000
p r ice
% w
ho b
uy
Revenue
0
2000
4000
6000
8000
10000
12000
14000
16000
0 500 1000 1500 2000 2500 3000price
© I.Popescu 2010 DPRM 3
Parametric estimate: log‐linear demand (Kili P3)
Log‐linear demand: d(p)= exp(a‐bp)Maximize: Revenue rate= p * exp(– 0.0017 p) ⇒ p*=$ 571
Log-linear wtp model
y = 1.4323e-0.0017x
R2 = 0.9406
0
0.2
0.4
0.6
0.8
1
1.2
0 500 1000 1500 2000 2500 3000price
% w
ho b
uy
Log-linear Demand
y = -0.0017x + 4.0228R2 = 0.9406
-1-0.5
00.5
11.5
22.5
33.5
4
0 500 1000 1500 2000 2500 3000
price
ln(d
eman
d)
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Graphical comparison
|ε|Quan
tity
|ε |
PriceQ
uan
tity
|ε|Quan
tityd=a-bp d=aebpd=apb
PricePrice
Linear Log-linearIso-elastic
a-bp a p−ε a exp(- b p)
log (D) = log(N) – b plog(D) = log(N) – ε log(p)D= N(1‐bp)
© I.Popescu 2010DPRM
Multi‐Pricing with Constraints –Generic Procedure
STEP 1: Forecast price‐response for each segment/product:– parametric model (regression ‐‐ using transaction or survey data)– non‐parametric (ok for setting a single price, or many prices w/o constraints)
STEP 2: Using the forecasts, set up a (usually non‐linear) optimization model (Solver) to determine optimal prices subject to various resource constraints (e.g. budget: EDM, LCC; capacity: Vertigo)
STEP 3: Set prices for each segment according to the model predictions (the logic: balance marginal revenues/profits per constrained resource unit)
STEP 4: If changing prices is feasible, go back periodically to STEP 1
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
London Congestion Charge
Revenue - Parametric versus nonParametric estimates
-200,000-100,000
-100,000200,000300,000400,000500,000600,000700,000800,000900,000
1,000,000
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Parametric Nonparametric
Parametric estimate ‐‐ uniform wtp
y = ‐0.0942x + 1.2924 R2 = 0.9323
0.00
0.25
0.50
0.75
1.00
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0
w
1‐F(w)
D(p)=N(1.29‐0.09p)
Objectives:1. max revenue2. minimize emissions s.t. revenue ≥ 500,000
R(p)=p N(1.29‐0.09p)
Speed(p) = 30‐.0625*D(p)/1000
Emissions(p)= 616.6‐16.7*speed (if <25)235.7‐1.4*speed (if >25)
© I.Popescu 2010 DPRM 3
Parametric vs. nonparametric estimates
• Parametric estimate
– Assume demand model defined by a modest number of parameters
– Estimate the parameters from data
• Demand is linear: D(p) = a ‐ bp
• Demand is N(μ , σ)– Pros: Concise description; can extrapolate beyond observed history; optimization
– Cons: Makes assumptions on form of response that may not be valid
• Nonparametric estimate
– Use the raw data directly to estimate demand without making assumptions on the functional form of the relationships
• Empirical histogram of demand volume
• Empirical histogram of reservation prices
– Pros: No assumptions; uses values actually observed
– Cons: Subject to “noise” in data; can’t extrapolate beyond history; hard to optimize
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Two broad strategies for estimation/forecasting• Bottom‐Up Forecasting
– Start with forming detailed models (“sub‐forecasts”)
• Individual customers; Segments; Locations; Channels
– Aggregate these sub‐forecasts into an overall forecast of demand
– Good at capturing detailed demand effects like difference in preferences across
segments, locations and channels
• Top‐Down Forecasting
– Start with high‐level model of aggregate demand (“super‐forecast”)
– Disaggregate this super‐forecast to form estimates of demand by channel, product,
location, etc.
– Good at capturing aggregate demand effects like seasonality, trends, etc.
Often in practice both strategies are used simultaneously
© I.Popescu 2010 DPRM 3
Extensions: Multi‐product demand & choice• So far customers faced a binary choice: to buy or not to buy. • What if they have multiple alternatives?
• Examples:– Multiple versions of a product (Vertigo part 3)
– Competing products/brands
– Choice of time periods
• Two approaches:
1. Multi‐product demand functionsThis can be estimated by regressing demand of each product against the
price of both products 1 and 2 (multiple regression)
D1(p) = a1 – b11 p1 + b12 p2 D2(p) = a2 + b21 p1 ‐ b22 p2
2. Discrete choice models based on utility max (multinomial logit ‐‐MNL)
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Resto menu prices
Menu A: €25 Menu B: € 35 Menu C: €40
© I.Popescu 2010 DPRM 3
Extremeness Aversion
Extremeness aversion (or Goldilocks pricing) : Add a high‐end version to your product line: people will trade‐up
Examples: restaurant menus, Carrefour champagne, electronics
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Online ad for the Economist a few years ago
• The Economist annual subscription options:
‐ Economist.com website only: $59‐ Print edition only: $125‐ Print edition PLUS website access: $125
16%0%84%
68%
32%
Asymmetric dominanceSegmentation: create value or the perception of value
© I.Popescu 2010 DPRM 3
Customer perception of prices
– Choice: extremeness aversion & asymmetric dominance– Framing & anchoring
– Reference prices & Endogenous expectations
– Fairness : prospect theory & dual entitlement
INSEADIoana Popescu
DPRM3
© I.Popescu 2010 DPRM 3
Summary – session 3
• Need to estimate parametric price response models (regression vs. raw/non‐parametric models) as input for price optimization (solver)
• Demand / price‐response estimation sources:
– Transaction data (most reliable)– Survey (willingness to pay, choice) data
• Different models will give different results: decision support only
• Combine data with judgment:– Understand consumer behavior (more to come)
© I.Popescu 2010 DPRM 3
Next time: markdown management
• The Retailer simulation (see website)
– read the case + task (in the course‐pack)– Understand the data file
– Play with the simulation ‐‐ available on all INSEAD desktops
– Prepare a group strategy
• Think how you would assess revenue impact (Bloomingdales)
• Groups will compete in class – there will be a Prize!
• Winner will present strategy to the class
• The simulation might not work on your laptop (see web for details …)