Product Adoption Across Categories in Emerging Markets: A ... Tewari - Category... · Final thought on Emerging Markets & Marketing • In this research, emerging market setting presented

Product Adoption Across Categories in Emerging Markets:

A Beckerian Framework

K. Sudhir and Ishani Tewari

Motivation

• Rapid growth in emerging markets creating a vast middle class for whom many categories of durables are becoming affordable

• Marketers need to understand drivers of category adoption to forecast sales growth and investment

• Our goal: Model category diffusion of durable goods in this dynamic environment

Modeling category adoption in an emerging market: Challenge #1

• Typically, diffusion models study the adoption behavior one category at a time

• Suitable in a developed country context, where invention drives availability and adoption

• Not often true in emerging markets

– As incomes rise, consumers can choose between many newly affordable categories simultaneously.

– Marketers need to think about nature of competition between categories

Modeling category adoption in an emerging market: Challenge #2

• Models need to be accommodate changing market environment of emerging markets

• Models based on household panel adoption data yield stable parameters of relative preferences for products

• May not be relevant in a highly dynamic environment where preferences towards products evolve rapidly

• We formulate a structural approach that characterizes why households choose one product before another – Value for a durable is derived from its downstream usage

– Embed this in a category choice model

Key Contributions

1. Developing a structural model for multi-category product diffusion in a dynamic environment

2. Leverage a household level panel dataset for China that has both multi-category adoption and time use data to estimate such a structural model

Modeling product usage: A “Beckerian” framework

• We model durables as time-saving technologies: usage reduces household chore time

• These time-savings:

– Can be either used to work or relax more – This tradeoff depend on household’s opportunity cost of time – Based on this, household allocates time allocation durable purchase

decision

• Follows Becker’s (1965) seminal model where utility maximizing

households trade off goods and leisure – Horsky (1990) conceptualizes durables in this way – Popular model to study female labor force participation

• We embed this Beckerian framework into a durable category

choice model – Train and McFadden (1976) : goods-leisure tradeoff and travel mode

choice

Some preliminary findings

1. We estimate effect of price, preferences and demographics on the choice across four popular HH durables

• Female labor force participation is an important moderator of utility for certain durables like fridge and washing machine

• Households with working spouse, especially in rural areas, have higher predicted ownership across all categories

2. Cross-category dependence: current ownership matters for future purchase

• Purchase of durable can vary depending on household’s current ownership combination

Plan for talk

1. Motivation and research questions

2. Our Approach

3. Preview of findings

4. Descriptives

5. Data: CHNS

6. Model and Estimation

7. Results

8. Counterfactual

9. Conclusion

DURABLE GOODS PENETRATION AND HOME PRODUCTION TRENDS (CHNS PANEL)

Durable goods penetration: China 1989-2011

Ownership by Income Bracket Income Bracket

Bottom Middle Top Bottom Middle Top

AC 0.08 0.13 0.21 Fan 0.68 0.74 0.77

Camera 0.05 0.09 0.18 Fridge 0.28 0.39 0.51

Car 0.03 0.03 0.06 Microwave 0.07 0.11 0.20

Cellphone 0.51 0.65 0.79 Motorbike 0.16 0.21 0.25

Computer 0.07 0.11 0.22 Sewing Machine 0.35 0.41 0.45

Cooker 0.48 0.56 0.64 Stereo 0.30 0.35 0.43

CTV 0.59 0.67 0.73 VCD 0.23 0.33 0.47

CTV 2 0.49 0.45 0.47 VCR 0.02 0.03 0.06

Washing Machine 0.43 0.52 0.64

Note: Income bracket defined by 3 percentiles of 2004 province-rural/urban income

• Also considerable variation across geographic and socio-demographic groups • Categories in highlights will be focus of analysis

Durable adoption in emerging markets especially suitable to be viewed through “Beckerian” lens

• Overall, Chinese spend more time in home production than in the US

Durable adoption in emerging markets especially suitable to be viewed through “Beckerian” lens

• Overall, Chinese spend more time in home production than in the US

• But, Chinese home production time is falling

• “Convergence” in US and Chinese time-use patterns

Heterogeneous Time Allocation

Time Allocation

cooking laundry grocery cleaning work leisure

Working Spouse 8.06 3.93 2.91 2.79 37.82 16.90

Non-Working Spouse 9.36 4.08 5.06 3.83 0.00 22.72

Big HH 3.56 1.25 2.34 1.27 39.28 19.81

Small HH 6.51 2.84 2.95 2.19 39.31 18.91

Below median income 3.85 1.40 2.41 1.35 44.97 18.53

Above median income 3.65 1.26 2.43 1.32 38.73 20.34

Urban 5.41 2.26 3.29 2.14 45.51 24.18

Rural 4.83 1.92 2.15 1.42 36.50 15.51

• Suggests opportunity cost of time may vary by observable HH characteristics

Data: Panel of Chinese households

• China Health and Nutrition Survey (CHNS)

• Household and individual level data

• Questions on durable ownership, prices time use, socio-demographics and preferences (especially related to food, physical activity and media)

• 9 waves between 1989-2011

• 9 provinces sampled

MODEL AND ESTIMATION

Households’ Utility

A household i derives utility from :

1

ijt it it ijt itU G L D

Goods they can buy with work income

Leisure they can enjoy

Extra income from durable ownership

Households’ Utility

A household i derives utility from : • Goods constraint

1

ijt it it ijt itU G L D

'   it it ijt ijtG V C p

1

J

ijt ijt it ijt

j

D O w W

Goods they can buy with work income

Leisure they can enjoy

Extra income from durable ownership

Total $ Choice of durable j

Price of j

Ownership of j

Wage rate*”extra”

work hours from durable

'

it it ijtV G D

Households’ Utility

A household i derives utility from : • Goods constraint

1

ijt it it ijt itU G L D

'   it it ijt ijtG V C p

'

1

J

it it ijt ijt

j

L L l O

1

J

ijt ijt it ijt

j

D O w W

Goods they can buy with work income

Leisure they can enjoy

Extra income from durable ownership

Total $ Choice of durable j

Price of j

Ownership of j

Wage rate*”extra” work hours from durable

“Extra” leisure hours from durable

• Leisure constraint

Generating a Multinomial Logit

• Consideration set excludes durables already owned • Main parameters--- α’s and β are further parametrized by socio-

demographic and preference variables

We get lognormal utility, with iid EV errors:

4

11

(1 ) ln( ) ln( ) 'J

ijt it ijt ijt it ijt ijt ijt it ijt ijtjj

u V C p L l O O w W

f

1 ...

0

1

Pr( ) Pr( )

exp( )

exp( )

ijt J I j

ijt ijt ikt ijt ijt ikt ikt

ijt

ikt i t

O

P u u j k f f j k

f

f f

Estimation: Two Steps

1. Estimate “time-boost” to household i’s leisure and work that accompany adoption of durable good j : Wijt and lijt

0 1ln

: Leisure (or Work time)

: Dummy for whether household owns durable or not

: Household fixed effect

ijt ijt i ijt

ijt

ijt

i

L Durable

L

Durable

2. Above estimates utilities and estimate via ML

1 2 3 4

1

( , , , , )j

i ijt ijt

i

L P Y

1 2 3 4arg max ( , , , , )i

i iL

RESULTS

Durable Ownership and Time Savings Ownership is generally correlated with more leisure and work hours

HH FE included, Robust SE in parenthesis *** p<0.01, ** p<0.05, * p<0.1 Specification in columns 5-8 is RE tobit

(1) (2) (3) (4) (5) (6) (7) (8)

ln Leisure ln Leisure ln Leisure ln Leisure ln Work ln Work ln Work ln Work

owncooker 0.166*** 0.382***

(0.03) (0.06)

ownwash 0.127*** 0.163***

(0.03) (0.06)

ownfr 0.0976*** 0.0397

(0.03) (0.06)

ownmb 0.0518* 0.828***

(0.03) (0.06)

Observations 12725 12726 12726 12725 13292 13293 13293 13292

Base: No Choice Model w/time-savings Estimate T-stat Price Constant 0.77 15.70 Working sp -2.29 -4.34 Urban indicator 0.47 0.89 Cooker Intercept Constant -3.28 -14.05 Non-traditional food pref 1.72 2.74 HH Size 0.04 0.55 Working sp 0.66 3.90 Urban indicator 0.16 0.78 WM intercept Constant -4.58 -13.08 HH Size 0.27 3.08 Piped water availability 0.35 1.44 Working sp 0.68 2.97 Urban indicator 0.35 1.27 Fridge intercept Constant -4.23 -8.06 HH Size 0.05 0.60 Fruits and vegetable preference 3.33 2.55 Working sp 0.51 2.34 Urban indicator -0.04 -0.15 Motorbike intercept Constant -4.86 -8.94 Household size 0.42 5.91 Physical activity preference -1.76 -1.22 Working sp 0.46 2.34 Urban indicator -0.79 -3.63 Observations 3909 Neg LLik 1.92E+03

MNL Results

Counterfactual: If we cut price of a durable by 10%, which households, based on what they currently own, are most likely to purchase?

current own combo % change in share when own price drop by 10%:

co,wm,fr,mb CO WM FR MB 0,0,0,0 0.0002 0.0012 0.0025 0.0004 0,0,0,1 0.0005 0.0005 0.0043 0 0,0,1,0 0.0006 0.0033 0 0.0009 0,0,1,1 0.0008 0.0005 0 0 0,1,0,0 0.0010 0 0.0048 0 0,1,0,1 0.0011 0 0.0055 0 0,1,1,0 0.0012 0 0 0.0002 0,1,1,1 0.0013 0 0 0 1,0,0,0 0 0.0073 0.0152 0.0021 1,0,0,1 0 0.0017 0.0145 0 1,0,1,0 0 0.0082 0 0.0021 1,0,1,1 0 0.0010 0 0 1,1,0,0 0 0 0.0185 0.0005 1,1,0,1 0 0 0.0088 0 1,1,1,0 0 0 0 0.0003 1,1,1,1 0 0 0 0

Average: 0.0004 0.0015 0.0046 0.0004

Conclusion • We develop a structural model for multi-category product diffusion in a

dynamic environment

• Leverage a household level panel dataset for China that has both multi-category adoption and time use data to estimate such a structural model

• We estimate effect of price, preferences and demographics on the choice

across four popular durable categories • Female labor force participation is an important moderator of utility for certain

durables like fridge and washing machine • Households with working spouse, especially in rural areas, have higher predicted

ownership across all categories

• Purchase of durable can vary depending on household’s current ownership

combination

Possible avenues of expansion • Dynamic model

Final thought on Emerging Markets & Marketing

• In this research, emerging market setting presented unique

opportunity to study cross-category product adoption – “Compressed” time line of diffusion and time-use change generates

temporal variation

– Wide variation in household incomes, social and cultural characteristics generates cross-sectional variation

– Household panel data

• More generally we can think of them as “fruit flies” for marketing

research

– Dynamics and heterogeneity allow researchers to test theories in a way which is not possible in developed markets.

• See “Emerging Markets” (Sudhir, Tewari et al. 2014) for review

THANK YOU!