retail notes 3

Description

Organize notes based on PPT and make them as concise as possible

Unformatted Attachment Preview

BU.450.740
Retail Analytics
Week 3
Pricing, Promotion, and Targeting:
Theory and Practice
Prof. Mitsukuni Nishida
Johns Hopkins Carey Business School
1
Announcement
• Feedback on Phase 1 report is available on canvas
• Homework 3 is due at the beginning of next lecture
2
Today’s Agenda
• Pillar 1: Advanced pricing at retailers
– Algorithmic pricing startups
– Pricing decisions at retailers
– Dealing with retail data
• Pillar 2: “Turbo” version of Microecon for pricing
– Firm’s behavior: Profit maximization, revenue and cost functions
• Applications: pricing new products, targeting strategy
– Equilibrium: Price competition in duopoly
• Pillar 3: Pricing analysis using retail data
– Elasticity estimation via Log-log price model
3
RETAIL NEWS:
How Instacart plans to gamify
omnichannel shopping
4
Instacart’s Smart Carts
• Instacart is trying to gamify grocery shopping with its Caper smart
carts, an executive said
• The carts, which contain touchscreens, can display games that
encourage customers to buy more
– Option to spin a virtual wheel on the smart cart screen for the
chance to win a prize.
– It could include giving a prize to a person who shopped three
times in the last two weeks, or encouraging people to build
bigger baskets
– “Using the cart’s location sensors and in-store navigation maps,
McIntosh said Instacart can envision giving shoppers a
challenge to find and scan certain products or types of products
in various aisles
– “Ultimately, where we want to take it is Pokémon Go,” McIntosh
said.”
5
“37-year-old quit Amazon and started 20
companies before coming up with
Instacart—now he’s worth $1.1 billion” (Sep
2023)
• Apoorva Mehta: 20 failed startups that he attempted, before finally
finding success with Instacart
– Over the next two years, his entrepreneurial attempts ranged
from an advertising startup for gaming companies to a social
network for lawyers. “Unfortunately, all of them failed,” said
Mehta.
• Finally, he finally turned to his own refrigerator for inspiration. Being
low on groceries “was an ongoing problem for me,” Mehta said, and
he figured he probably wasn’t alone
• He also saw a gap in a thriving online delivery market. “This was
2012 and we were ordering everything online, except for groceries,”
said Mehta. “I decided I was going to change that. I started coding
the first version of the Instacart app and, three weeks later,
6
Instacart was born.”
PILLAR 1: ALGORITHMIC
PRICING STARTUPS
7
Startups help retailers optimize pricing
Source: CBInsights
“Startups are leveraging AI and IoT technology, among others, to
help retailers sell to customers at discounted or optimized price
8
points” (Reddy, 2019)
9
PILLAR 1: PRICING
DECISIONS AT RETAILER
10
Hierarchy of Pricing Decisions
From aggregated to disaggregated
• Store (e.g., EDLP vs. Hi/Lo)
• Category (e.g., Softdrink, meat, …)
• Product (e.g., Minute Maid 12 oz. OJ)
• Individual consumer (e.g., promotion via emails/apps)
We utilize different methods to answer pricing in different level
11
EDLP vs. Hi/Lo
• Everyday low pricing (EDLP)
– Doesn’t mean “lowest” price
– Reduces advertising/operating expenses
– Reduces stockouts & improves inventory management
• High/low (or “promotional”)
– Frequently discount initial prices
– Creates excitement
– Sells slow-moving merchandise
12
EDLP vs. Hi/Lo: Empirical Patterns
Pricing strategies are made at the store level, not the chain level
=> importance of considering (1) competition and (2) demographics
Method: The authors asked individual store
managers to choose which of the following
categories best described their store’s pricing
policy:
• Everyday Low Price (EDLP): Little reliance on
promotional pricing strategies such as temporary
price cuts. Prices are consistently low across the
board, throughout all packaged food departments
• Promotional (Hi-Lo) Pricing: Heavy use of
specials, usually through manufacturer price breaks
or special deals.
• Hybrid EDLP/Hi-Lo: Combination of EDLP and HiLo pricing strategies.
For (1), supermarkets in the neighborhood tend to coordinate on pricing strategies
For (2), EDLP is often used at lower income consumers with larger families in
13
urbanized areas
Source: Ellickson and Misra (2008)
PILLAR 1: DEALING
WITH RETAIL DATA
14
Where Do We Get In-store
Data?
• Where do the data originate?
• Source 1: TLOG, Some firms capture this data themselves
– Example: scan of weekly features
– More often this is an archival process rather than a database
for analysis
– Only available for the retailer
• Source 2: Scanner data
– Syndicated data companies
– Available for a sample of stores
15
Syndicated Data Provides in Retail
• Nielsen (https://www.nielsen.com/)
• IRi (https://www.iriworldwide.com/en-US/)
• NPD group (https://www.npd.com/)
16
Traditional Retail Data Begin at
Cash Register
• Transaction log data (TLOG)
– Captured via scanner at check-out register of every retail store
– Primary source of data for all retailers
• Data look like retail store receipts
17
TLOG provides a detailed
view of what happened
• Description of item that was sold
• Volume purchased
– Number of units
– Weight
• Price paid
– Price per unit
– Price per unit volume
• Total amount paid for basket of items
– Gross Sales, Tax, Net Sales
• Customer ID
– Frequent shopper card
– Credit card number
18
What TLOG Does not Record:
In-store Marketing
• Imagine a customer bought Progresso soup, which was offered at
4 for $7 at Giant Food in Baltimore
• Missing Data include …
– Progresso
• Regular price? Other promotions? Television advertising?
• In-store signage? Feature advertising?
– Competing Brands
• Campbell’s soup price? Promotion? Advertising?
• Campbell’s in-store signage? feature advertisement?
– Retail Competition
• Pricing in other supermarket stores in the neighborhood?
• Information above will help us better understand why Progresso
19
was purchased
TLOG/Scanner Data Need to Be
Heavily Processed before It Is Usable
Raw data are never usable
As a data scientist or brand manager, you will work with highly
processed data after
1. Error correction
2. Aggregation
20
Error (1): Missing Observations
Units Sold of Minute Maid 12oz Frozen OJ in Chicago Dominick’s
Stores
Bad Data: missing unit sales for several weeks
21
Error (2): Errors in Consumer Scanning
Einav et al. (2010) find
• Accuracy of Homescan data are in line with other (government) economic data
• Quantities purchased are reported more accurately in Homescan than are prices
• Nielsen uses store-level prices as estimate of what household actually paid. But in
reality retailers have multiple prices due to loyalty card, temporary price reduction,
& shopper specific price promotions
Level of Aggregation
• We aggregate retail transaction in four dimensions
1.
2.
3.
4.
Product
Geography
Time
Customer
23
1. Product: We Aggregate UPCs (or
SKUs) into Products
Store Brand 12 oz OJ
• Regular
• With Pulp
• With Calcium
24
Remark: Stock-Keeping Unit and Universal Product
Code
25
2. Geography: We May also Want
to Aggregate across Geographies
Dominick’s has 96 stores in Chicago Metro Area
Level of Aggregation
could be:
• All stores
• Price Tiers
• Zones
• Single store
26
3. Time: Data Are Collected Every
Second, but…
• …we aggregate data over time periods, such as day, week,
month, or year
• Managerial decisions decide time period
– For many retail stores, pricing decisions are made weekly
– For strategic planning decisions, one may look at quarterly or
annual time periods
27
4. Customer: Aggregation by
Customer Facilitates New Analysis
• By aggregating purchases for a shopper, we can conduct
– Targeted promotions
– Customer lifetime value (CLV) analysis
• Targeted promotions: Segmentation based on (1) recent purchase
behavior (e.g., baby food) or (2) customer characteristics (e.g.,
age, income, married, # of children)
• CLV analysis: Analysis based on shopper identification, which is
one of the keys to retail differentiation
28
Summary: Principles of Aggregation
• To facilitate analysis and decision making, we aggregate
observations
• When aggregating, we need to be careful not to lose valuable
information
– e.g., price and quantity variations over time
• Strategy?
– We start with very aggregate data, and then move toward finer
cuts of the data
– Example
• Start: All Frozen OJ, All stores in Chicago
• Next: Specific brands, brand sizes, specific geographies
29
Summary (cont’d): Model to Match
Degree of Aggregation
• Market Data: Very aggregated
– If units sold are typically large (units > 10), we often use a loglog regression model (this week)
– If units sold is a small number, such as 0, 1, 2, etc., we often
use a poisson regression, logit, & probit models
• Customer-level or Firm-level Data: Disaggregated
– If we look at customer data (e.g., Buy vs. Not Buy) or firm-level
data (e.g., Enter vs. Not Enter), we use a logistic regression
model (next week)
30
Emerging Sources of Retail Data
• Online Store
– Click stream: Search & purchase behavior
– Customer feedback: Customer product reviews
• Email
– Allows for testing and targeting
• Product Reviews
– Faster reaction to problems
• Social Media
– Communication, word of mouth
– Problem resolution
– Shopper behavior
• Search platforms: Google, ChatGPT
31
Digression: Product Reviewers Are Predictable
Are some reviewers better (or worse) than others at identifying
successful new products?
According to Kellogg Research Study,
• Reviewers themselves are amazingly consistent
– If a reviewer recommends a winner today, she/he will tend to
recommend a winner tomorrow
– If a reviewer recommends a loser today, she/he will tend to
recommend a loser tomorrow
32
PILLAR 2: SUPPLY: FIRM’S
DECISION MAKING
33
Profits, Revenue, Cost
π(Q) = TR(Q) – TC(Q) Q: quantity that a firm produces/supplies
π(Q): Total profits
TR(Q) = Total revenue = (Price) * (Quantity) = P*Q
TC(Q) = Total cost = (Fixed cost) + (Variable cost) = FC + VC(Q)
•
MR = dTR(Q)/dQ
– MR (Q): Marginal revenue = additional revenue from selling one more unit
• MC = dTC(Q)/dQ
–
MC (Q): Marginal cost = additional cost from producing one more unit
Note: formula for taking the first derivative of f(Q) = Q^b with respect to Q
• df(Q)/dQ = b* Q^(b-1).
• If b =1, df(Q)/dQ = 1. If b=0 (i.e., f(Q) = 1), df(Q)/dQ =0
34
Principle of Profit Maximization
• Principle of Maximization: Take derivative of objective function and
set equal to 0 (“First-Order Condition”)
– Profit maximization ⇛ find Q to set Marginal profits = MR-MC= 0
– Revenue (TR) maximization ⇛ find Q that sets MR= 0
• To see this, remember
(Q) = TR(Q) – TC(Q)
TR: total revenue, TC: total (economic) cost
• To find the Q that maximizes (Q), we find Q such that
dTR
dTC
d
=
–
=0
dQ
dQ
dQ
MR – MC = 0
MR = MC
• Intuition
– If MR > MC  Profit can be increased by producing more
– If MR < MC  Profit can be increased by 35 producing less Profit Maximization for Monopolist Profit = π = TR – TC P TR: Total revenue = P*Q TC: Total cost (= FC + VC) P* To find the profit-maximizing quantity, set the first derivative of π with respect to Q = 0:  TR TC = − =0 Q Q Q MR MC π Q* Q MR MC Goal: To maximize the profits, we find Q such that MR = MC Q* 36 Q APPLICATION OF DEMAND AND SUPPLY ANALYSIS (1): PRICING NEW PRODUCT/SERVICE 37 38 Mini case: Space travel • Suppose (1) the projected demand curve for airspace travel is P = $400K– 150Q, (2) marginal cost (MC) is $100K per person, and (3) fixed costs for the company are $100M • Questions: Profit maximizing price and quantity? Profits? • Answer: MR = d(TR)/dQ = d(P*Q)/dQ = d{(400K-150Q)*Q }/dQ = d(400K Q – 150Q^2)/dQ = 400K – 300Q • So setting MR = MC (=$100K) gives Q = 1000, P = $250K, Profits = Rev – Costs = PQ – $100KQ - $100M = $250M - $100M - $100M = $50M 39 APPLICATION OF DEMAND AND SUPPLY ANALYSIS (2): TARGETING 40 3 Potential Segments of Buyers • Consider the following 3 segments of buyers for my company’s pipe products Recap: P < (B – B’) + P’ = Maximum price my firm can set (“WTP”) Differentiation Reference Value Price 41 Map Segments as Demand Curve 42 If marginal cost is $65, target the toxic materials segment only If selling to Toxic segment only, Profits = ($100 - $65) * 35 = $1,225 If selling to Toxic & Sprinkler segments, Profits = ($80 - $65) * 50 = $750 < $1,225 43 If marginal cost is $30, target the toxic materials & sprinkler segments If selling to Toxic segments only, Profits = ($100 - $30) * 35 = $2,450 If selling to Toxic & Sprinkler segments, Profits = ($80 - $30) * 50 = $2,500 If selling to all segments, Profits = ($40 - $30) * 100 = $1,000 < $2,500 44 Microecon Theorey at Work: Find Q such that MR = MC for Profit-Maximization Marginal Cost (MC) = $65 Marginal Cost (MC) = $30 Demand Profit-maximizing Q when MC=30 45 Marginal Revenue (MR) Profit-maximizing Q when MC=65 COMPETITION IN PRICE: DUOPOLY 46 Market Structure: A Spectrum Market structure • Provides predictions on prices and quantities • Tells when profits converge to zero Large # of firms No market power Few firms Some market power Only one firm Full market power Perfect Monopolistic Oligopoly competition Competition “price taker” “strategic interaction” Today Monopoly “price setter” 47 Bertrand Duopoly with Homogeneous Product • Two firms produce identical products– bottle of water • Firms simultaneously choose price levels P1 and P2 • One-shot static game • Resulting price determines the quantity Q • Each firm – takes the other firm’s price as given – chooses the price that maximizes its profits 48 Bertrand Equilibrium: Definition • Bertrand equilibrium is a pair of prices (P*1,P*2) such that – P*1 is firm 1’s best response to P*2 – P*2 is firm 2’s best response to P*1 • In other words, this equilibrium is a pair of (P*1,P*2) such that no firm can increase profits by unilaterally changing its price 49 Bertrand Duopoly: Bottle Water Example • Suppose both firms have total costs $10*Q, where a unit is a case of bottles => Marginal cost (MC) is $10
• Linear demand curve: QT = 100 – P
• Only possible equilibrium is P1 = P2 = MC = $10
• Why? To see this, we first construct profit function
– Profits for firm1: Π1 = P1 * Q1 -10 Q1 = (P1 -10) Q1
– Profits for firm2: Π2 = P2 * Q2 -10 Q2 = (P2 -10) Q2
50
Bertrand Duopoly: Bottle Water Example
• Why is (P1,P2) = (10, 10) Bertrand equilibrium?
• Suppose P1 < MC = $10 – No production because profit is negative • Suppose P1 > MC = $10
– Say P1 = 12. If P1 = P2 ,Q = 88, Q1 = Q2 = 44
– Π1 = (P1 – MC) * Q1 = (12 – 10)*44 = 2* 44 = 88
– If slightly undercut the rival by setting P1 = 11.99
• Π1 = (P1 – MC) * Q1 = (11.99–10)*88.01 =1.99 *88.01 = 175
– Is (P1, P2) = (11.99, 12) an equilibrium? No, because Firm
2 has an incentive to deviate by slightly undercutting
• Suppose P1 = P2 = MC = $10
– No incentive to deviate => Bertrand equilibrium
51
Bertrand Duopoly: Properties
• Driver of results: When products are homogeneous, undercutting
firm will take the all demand in price competition
• “Aggressive” behavior by one firm (price cutting) is met by
“aggressive” behavior by rivals (price cutting).
• Similarly, price increase (passive behavior) is met by price increase
• As # of firms goes from 1 to 2, equilibrium price goes from
monopoly price to perfect competition price
– Quite strong result
– In real-world industries with a few firms
• Increase in # of firms normally implies a gradual decrease in price
• Firms seem to make more than zero profit
• We call the above result “Bertrand’s paradox”
• This paradox is resolved when we introduce product differentiation
52
(reprise) Amazon’s competitive price
matching to rivals’ price +/-
Source: Chen, Mislove, and Wilson (2016)
53
APPENDIX: KEY TERMS
54
Key terms
•
Everyday low-pricing (EDLP) strategy: A pricing strategy that
stresses continuity of retail prices at a level somewhere between
the regular nonsale price and the deep-discount sale price of the
retailer’s competitors.
•
Geofencing: Offering localized promotions for retailers in close
proximity to the customer, as determined by phone location
technology.
•
High/low pricing strategy: A strategy in which retailers offer prices
that are sometimes above their competition’s everyday low price,
but they use advertising to promote frequent sales.
•
Stock-keeping unit (SKU): The smallest unit available for keeping
inventory control. In soft goods merchandise, an SKU usually
means a size, color, and style.
•
Universal Product Code (UPC): A barcode symbology that is widely
used in the United States, Canada, UK, Australia, New Zealand, in
55
Europe and other countries for tracking trade items in stores.
(end of slides)
BU.450.740
Retail Analytics
Week 3
Log-log model in R
Prof. Mitsukuni Nishida
Johns Hopkins Carey Business School
1
Today’s Agenda for Pillar 3
• Technical issues when interpreting outcomes
– Omitted variable bias
– Fit of the model
– Hypothesis testing: F-statistic
• Log-log pricing model in R
– Concept: Estimate price elasticity
– Application: Energy bar promotion decision
2
Preparation
• Open RStudio
• If R loads workspace from /.Rdata, clear the workspace, such as
loaded data, scripts, etc
– rm(list=ls())
• Load data
– load(“marketing.rda”)
• Set the working directory by
– setwd(“C:/Users/mnishid2/Dropbox/teaching/2023_24/Retail
Analytics/Coding in R/w3/Omitted_variable_bias”)
• Load required R package “tidyverse” for easy data manipulation and
visualization
– install.packages(“tidyverse”)
– library(tidyverse)
3
INTERPRETING OUTCOMES
(1): OMITTED VARIABLE BIAS
4
Recap: 3 Assumptions for Unbiasedness
1. Linear in parameters
The model in population is a linear combination of explanatory
variables
2. Random sampling
We have a random sample of N observations of (Y, X)
3. Zero conditional mean
The error epsilon has an expected value of zero given X
– Condition 3 says “X variables are exogenous such that X
variables do not correlate with epsilon (error term)”
– This condition can be violated in several ways
• For instance, when key variables to explain Y are omitted from the
estimation equation & those variables interact with X (“omitted
variable bias”)
5
RECAP: Simple Linear Regression
• We construct a linear regression model
Yi =  0 + 1 X 1 + 
– Where Y: sales, X1: youtube ads expenditure
• We use lm() to run a regression
– lm(sales ~ youtube, data = marketing)
Y (“regressand”)
X (“regressor”)
6
RECAP: Multivariate Regression: Outcome
• We run
– reg_sales_all
Purchase answer to see full
attachment