ECONOMICANALYSIS AND APPLICATION OF BIG DATA

(1)

ECONOMIC ANALYSIS AND APPLICATION OF BIG DATA Monique S.K. Wan Department of Economics

(2)

Agenda

• Definition of Big Data

• Industrial Revolution 4.0

• Applications of Big Data

– Banking and risk management

– Marketing and recommendation system – Government and smart city

• Implications of Big Data

– Business landscape and job nature – Data as asset

– General Data Protection Regulation – Role of economists

(3)

Definition of Big Data

“Big data is high-volume, high-velocity and high-variety information assets that demand cost-effective, innovative forms of information processing for enhanced

insight and decision making. ”

– Doug Laney of Gartner, Inc.

(4)

Every minute:

• 204 million emails

• 1,800,000 Facebook

• likes

• 72 hours of videos upload to YouTube

(5)

Broader Definition

• Veracity

⁴

– Quality of the data

(6)

Broader Definition

• Valence

⁵

– How big data can bond with each other?

• Measure the ratio of actually connected data items to the possible number of

connections that could occur within the collection

(7)

Broader Definition

• Value

⁶

– How can big data benefit your organization?

• Require clear business

strategy and data analytics tools

• Cross Industry Standard Process for Data mining (CRISP-DM; Shearer 2000)

Source: Provost and Fawcett (2013)

(8)

Industrial Revolution 4.0

Source: NetObjex Blog

(9)

How Big is Big?

• Company Signet Bank

– A small proportion of customers actually account for more than 100% of a bank’s profit from credit card operations

• Business objective

– Model profitability – Not just default probability, but also pricing

• Data understanding/ Data preparation

– Use of algorithms to learn the willingness-to-pay and charge-offs – Acquire information via 45,000 scientific tests

– Increase in bad accounts (charge-offs)  Cost-and-benefit analysis

(10)

How Big is Big?

• Deployment/ Action

– Offer different customers different rates, terms, offers

• Business value

– A new spin-off in 1994 (Capital One)

• Richard Fairbank and Nigel Morris

– Information-based strategy – Lowest default rate

– Largest credit card issuers

(11)

How Fast is Fast?

• Vehicle with 100 telemetry sensors that capture data to improve track performance

• Data science team

– Driver

– Team principal/Race

engineers/Operations room analysts/trackside analysts

• Real-time data analysis

– Lap/split times – Tire/brake temp – Air pressure – Air flow

– Engine performance – GPS information

(12)

How complex can it be?

• Business understanding

– Provide streaming movie and TV service

• One-third of peak-time Internet traffic in US

• 65M members in over 50 countries

• 100M hours of TV shows and moves a day

– Build recommendation engines

• Data understanding / Data preparation – Unstructured data

• 80,000 features (costly!!!)  Cost-and-benefit analysis

– Existing data = {Customer ID, movie ID, ratings and the date the movie was watched,…}

– New data = {Time spent on selecting movies, how often playback is stopped, tags of the movies,…}

(13)

• Modeling

– Measure similarity among products – Measure similarity among customers

– Predictive analytics make recommendation

• Improve user experience

• Induce to consume more

• Additional business value

– Product Innovation

• Develop new business as content creator

– Outbid HBO: House of Cards directed by David Fincher and starred by Kevin Spacey

Valuable ?

(14)

Banking & Risk Management

• Business question

– What is the likelihood of default for this loan applicant ?

• Data collection

– Past loan records: Applicant’s profiles and outcome (default or not default)

– The profile of this loan applicant

• Modeling (supervised models)

– Classification tree – Logistic regression

– Support vector machines – Neural networks

(15)

Decision Tree For Classification

• Business decision

– Should we write off the loan?

• Model

– Write off the loan if

• Not employed; and

• Remaining loan balance is huge; and

• Older than 45

(16)

Decision Tree

• Measurement of Purity

– Error rate

• Definition: 𝐸 = 1 − 𝑚𝑎𝑥 𝑝₊, 𝑝₋

• Example: 𝐸 = 1 − ⁹

10 = 0.1

– Gini index

• A measure of total variance across the 2 classes

• Definition: 𝐺 = 𝑝₊ 1 − 𝑝₊ + 𝑝₋ 1 − 𝑝₋

• Example: 𝐺 = ₁₀⁹ ₁₀¹ + ¹

10 9

10 = 0.18

– Entropy

• Definition: 𝐷 = −𝑝₊ log₂ 𝑝₊ − 𝑝₋ 𝑙𝑜𝑔₂ 𝑝₋

• Example: 𝐷 = − ⁹

10𝑙𝑜𝑔₂ ⁹

10 − ¹

10𝑙𝑜𝑔₂ ¹

10

(17)

Logistic Regression

Target variable

𝑦_𝑖 = ቊ1 Write − off

0 Not Write − off

Feature variables

• 𝐴𝑔𝑒_𝑖

• 𝐵𝑎𝑙𝑎𝑛𝑐𝑒_𝑖 𝒍𝒐𝒈𝒊𝒕 𝒚_𝒊 = −𝟔𝟎 + 𝑨𝒈𝒆_𝒊 + 𝟏. 𝟓𝑩𝒂𝒍𝒂𝒏𝒄𝒆_𝒊

(18)

Forecasting Recession

Target variable

𝑦_𝑡 = ቊ1 Recession 0 Otherwise

Feature variable

𝑌𝑖𝑒𝑙𝑑 𝑐𝑢𝑟𝑣𝑒_𝑡 = 𝐿𝑅 − 𝑆𝑅 𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑟𝑎𝑡𝑒

(19)

Marketing Strategy

• Domain Knowledge

– Cost = Contact Hours × Wage Rate = $25 – Product Price = $200

– Product Cost = $100

– Respond (R)  Benefit = $200 − $100 − $25 = $75 – No Respond (NR)  Benefit = −$25

• Expected benefit of targeting

𝐸𝑉 = 𝑝_𝑅 × 𝑣_𝑅 + 𝑝_𝑁𝑅 × 𝑣_𝑁𝑅

𝐸𝑉 = 𝑝_𝑅 × $75 + 1 − 𝑝_𝑅 × (−$25) > 0 𝑝_𝑅 > 0.25

• Business action

– Target the customers if their estimated probability of responding to the marketing plan >25%

(20)

Data Analytics

Model Prediction

(Probability Score) Action Class

0.87 Contact Respond

0.65 Contact Not Respond

0.24 Do Not Contact Respond

⋮ ⋮ ⋮

(21)

Recommendation System

• Personalization

– Predict user’s interests and recommend products/

services

• Spotify, YouTube, Netflix

• Data types

– Characteristic information

• Users: User’s background information, Preferences

• Items: Categories, Specific features

– User-item interactions

• Purchase histories, User ratings, Search engine queries, Browsing sequences

(22)

Recommendation System

• Content-based system

– Based on user’s profile features and item features

• New products falling outside the past “interests” will not be recommended

• New customers with not much information may be ignored

• Collaborative filtering system

– Utilize user-item interaction

• If a user likes item A, and another user likes A and B.

Then, B will be recommended to first user

– Identify clusters of users and items

(23)

Types of Recommendation System

Content-Based Collaborative Filtering

Source: https://www.youtube.com/watch?v=Eeg1DEeWUjA

(24)

Government & Smart City

• Technology

– More information with higher accuracy and faster information flow

• Applications

– Transportation: Bus scheduling/ smart parking – Banking: Cashless and contactless payments

– Waste/ Drainage management system: Robotics and assistive technologies

– Energy conservation: Data-empowered urban environments

– Healthcare: Technology-enabled homes

(25)

Government & Smart City

• Chakravorti and Chaturvedi (2017)

– Citizens/People Components

• Inclusivity, environment and quality of life, state of talent and the human condition, talent development

– Economy Components

• Global connectedness, economic robustness, entrepreneurial ecosystem, innovation capacity.

– Institutions Components

• Freedoms offline and online, trust, safety and security, public services

(26)

(27)

Business Landscape & Job Nature

Business Models

• New business/ business models

– Data consulting companies – Cloud storage

– Apps developers – Automation

– Sharing economy

• Traditional business models

– Entertainment companies – Media

– Intermediaries

Job Nature

• Job creation

– Data analysts

– Computer scientists

– Decision makers with strong domain knowledge

• Job displacement

– Old fashioned sales representatives

– Jobs with routine tasks

(28)

Cloud and Data as Asset

Alibaba-Tencent struggle disrupts, invigorates China internet

Source: https://www.bnext.com.tw/article/44785/shun-feng-with-alibaba-china-most-strong-logistics-logistics-platform-of-decorum Source: http://fortune.com/longform/alibaba-tencent-china-internet/

(29)

Produce and Manage Data Assets

“

Infonomics is the concept that

information is, or should be, an actual enterprise asset.

^”

– Doug Laney

(30)

Global Data Protection Regulation

• 7 Principles

– Purpose limitation – Data minimization – Accuracy

– Storage limitation

– Lawfulness, fairness and transparency

• Requiring the consent of subjects for data processing

– Integrity and confidentiality

• Anonymizing collected data to protect privacy

– Accountability

• Safely handling the transfer of data across borders

• Providing data breach notifications

(31)

Role of Economists

• Macroeconomists

• Microeconomists

• Econometricians

James Heckman Daniel McFadden John Nash Daniel Kahneman Robert Lucas Robert Solow

(32)

Jobs for Economists

• Examples of tech companies that have hired PhD Economists

Pat Bajari Hal Varian Susan Athey Johnathan Hall

(33)

Role of Economists in Tech Firms

• Empirical Industrial Organization (Athey and Luca, 2019; Shum, 2016)

– Estimate demand function and market power

– Design online advertising strategies and estimate returns to advertising

– Design review and reputation systems and analyze the effect of reviews

– Evaluate acquisitions, exclusive deals and strategy – Promote incentives in marketplaces

(34)

Demand Function

• Industrial organization

– More about supply-side (firm-side)

– How much market power do firms have?

• Market power

– Markup:

^{𝑝−𝑚𝑐}

𝑝

– Marginal cost (𝑚𝑐) is unobserved!

• Observation: High price (𝑝) in an industry

– High market power?

– High 𝑚𝑐?

(35)

Demand Function

• Monopoly

𝐦𝐚𝐱

𝒑

𝒑𝒒 𝒑 − 𝑪 𝒒 𝒑

• First-order condition

Price elasticity

of demand

(36)

Traditional Approach to Demand Estimation

• Consumer demand as a utility maximization problem

max𝑥₁,𝑥₂ 𝑈 𝑥₁, 𝑥₂ s.t. 𝑝₁𝑥₁ + 𝑝₂𝑥₂ = 𝑀

– 𝑝₁ and 𝑝₂ are prices of good 1 and good 2 – 𝑀 is income

• Solution

– 𝑥₁^∗ 𝑝₁, 𝑝₂, 𝑀 and 𝑥₂^∗ 𝑝₁, 𝑝₂, 𝑀

• Question

– Do we really need 𝑥₁^∗ good 1 and 𝑥₂^∗ good 2?  Discrete choice

(37)

Discrete Choice Modeling

• Product nature/ Consumption pattern – Many alternatives, too many parameters

• Automobile, airlines, cereals, toothpaste

– Consumers only choose one of the available options (discrete choice)

_𝑖𝑗

is utility shock that is not observed by others

• Consumer preference

– Choose product 𝑗 when 𝑈

_𝑖,𝑗^∗

> 𝑈

_𝑖,𝑘^∗

for 𝑘 ≠ 𝑗

• Objective

– Estimate the choice probabilities for all goods

(39)

Economists at Amazon: Making an Impact

(40)

References

• Athey (2017). Beyond prediction: Using big data for policy problems. Science 355:

483-485.

• Athey and Luca (2019). Economists (and economics) in tech companies. Journal of Economic Perspectives 33(1): 209-230.

• Varian H.R. (2014). Big data: New tricks for econometrics. Journal of Economic Perspectives 28(2): 3-28.

• Rudebusch G.D., and Williams J. (2005). Forecasting recessions: The puzzle of the enduring power of the yield curve. Journal of Business & Economic Statistics 27(4):

492-503.

• Chakravorti B. and Chaturvedi R.S. (2017) The “smart society” of the future doesn’t look like science fiction. Harvard Business Review.

• Provost F. and T. Fawcett (2013). Data Science for Business. 1st Ed. O’ Reilly Media.

• Bernard M. (2016). Big Data in Practice: How 45 Successful Companies Used Big Data Analytics to Deliver Extraordinary Results. 1st Ed. Wiley.

• Shum M. (2016). Econometric Models for Industrial Organization. World Scientific Lecture Notes in Economics Vol. 3. World Scientific.