PERC_BIG Data CreditScoring_102015_1(2)
Post on 14-Apr-2017
134 Views
Preview:
Transcript
Credit Scoring
1
Credit Scoring & “Big Data”
PERC Presentation: Dr. Michael A. TurnerOctober 26th, 2015Credit Reporting in Asia-Pacific and Personal Data ProtectionXi’an, Peoples Republic of China
Select PERC Supporters Include…Foundations& Nonprofits
Government & Multilaterals
Trade Associations
Private Organizations
2
Our Footprint
AfricaCameroonKenyaSouth AfricaTanzania
North America/ CaribbeanCanadaMexicoTrinidad & TobagoUnited States of America
AsiaBruneiChinaHong KongIndiaIndonesiaJapanMalaysiaPhilippinesSingaporeSri LankaThailand
Australia/OceaniaAustraliaNew Zealand
EuropeFrance
Central/South AmericaBoliviaBrazilChileColombiaGuatemalaHonduras 3
PERC’s Alternative
DataInitiative
(ADI)PERC advocates the inclusion of alternative data for use in credit grantingalternative = regular bill payment data from telecoms, energy utilities, rental payments and other such non-financial services that are valuable inputs for credit decisions
Research Consensus Confirms Benefits of Alternative Data
5
March 2015
Many Organizations Examined Alternative Data
• PERC• CFSI• Brookings Institution• Boston Fed• World Bank• IFC• PBOC CRC• Privacy
Commission (AUS, NZ, EU)• Equifax• Experian• VantageScore• FICO• Lexis-Nexis• MicroBilt• SAS Institute
Types of Data Examined: Utility payments, Rent Payments, Telecom Payments, Pay TV, Cable, and Underutilized Public Records
7
Other Alternative Data Being Used
Rental data United States (certain locations) Colombia (in Bogota area) South Africa (Johannesburg area)
Trade supply (not trade credit) for FMCG
Agricultural supply data (for rural lending)
Some fit into credit bureau model, others do not
Just what is “Big Data” anyway?
“Even with decades of this data, what would we know?”
• More marketing term than discipline• Impetus was harnessing large volumes of data
from http servers (make sense of users of e-commerce and social media)
• Unstructured and interpreted rather than specific answers to specific questions (interpret people’s intentions by analyzing their clickstreams)
Credit Reporting is Ripe for a Disruption
Updates take at least 30 days.Blind to critical differences among borrowers (deadbeats vs. victims of business cycle)
Their data is old. Their approach archaic and shrouded in mystery. ”
“All data is credit
data.”
Big Data Meets Underwriting
Updates take at least 30 days.Blind to critical differences among borrowers (deadbeats vs. victims of business cycle)
Big Data Start Ups
The future is already here in credit risk assessment
12
Big Data Caveat hic
esse draconum
While “Big Data” holds forth great promise, it carries significant near term risks:• Misplaced faith• Data Fiefdoms • Investments misallocated away from “trenches” to what’s “trendy”
Big Data Leads to BIG Failures
Spurious Correlations
Big Data Explains Everything And Nothing
Big Data uses more false informationNoise overwhelms the signal in large data setsWidespread overfitting of data with model
The single most predictive variable of credit risk is “dishwasher ownership.”
Big Data Entirely
RemovesThe
ConsumerConsumers would be overwhelmed by use of thousands of variables in credit grantingOECD Fair information Practices place consumer at center of data sharing regime. Big Data would severely challenge Notice, Choice, Access, Redress
Big Data & Data
Fiefdoms
Obsession with Big Data may counter-intuitively result in reduced data access
18
Big DataSome final thoughts:
Approach with EXTREME caution Big Data is a complement to, not substitute for established
predictive data Dishwashers aren’t causally related to credit risk
Discourage Data Fiefdoms› Prohibit conflicts of interest in ownership structures› Encourage reasonable access to vital data sets
Keep Consumers at the Center Big Data removes consumer Big Data threatens privacy
Leap into the Trenches› Invest in efforts to digitize “traditional” alternative data› Encourage use of established predictive data in origination process
6409 Fayetteville Road Suite 120-240Durham, NC 27713www.perc.net(919) 338-2798 x803
top related