Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Global Forum Turkey 2015

No description
by

Creditinfo Lietuva

on 11 September 2015

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Global Forum Turkey 2015

Innovations in Predictive Modelling (Unusual Scorecards)
Big Data
Volume
Veracity
Velocity
Variety
V
V
V
V
Huge data
size - typically
terabytes, petabytes +
High speed
of data flow,
change and processing
Many types
of sources -
tables, texts, music, photos, videos, ...

Different data
quality level,
uncertainty and reliability
Credit Bureau
Enterprises
Income
Insolvency
Court
Arrests
Bailiffs
Taxes
Bankruptcy
Mortgage
Vehicles
...
What makes data Big?
Traditional data sources
Traditional way
#3 create solutions and services for credit risk evaluation
Alternative Big data sources
Social media
Social networks
Psychometrics
Digital footprints
Bank transactions history
Telcos behavior
Utility behavior
...
blank profile
personal
profile
Alternative data models
examples of Creditinfo experience
Undeclared electricity
consumption and thievery
Credit risk modelling using
social network data
1 million+ profiles analyzed
~10 000 data points per person
5 TB+ data collected
Alternative scoring
Scoring with
social network data
Blended scoring
for SME's
#1 establish standard connections
to public sources
#2 combine public and private data
Social network data
Use cases
Employee control
Collecting info about potential employees / tracking employee social behaviour

Customer attraction
Increasing company brand awareness through influence of influential people

Credit scoring
Using personal social network data to make credit scoring models
Social scoring
Petabytes of data are saved every day worldwide (petabyte = ~1 000 000 GB)

Properly used = significant competitive edge in the market

Alternative data - excellent source for
tailor-made solutions

Mantas Tartėnas
Nataliya Soldatyuk

The first thing is character <...> Money cannot buy it...
Because a man I do not trust
could not get money from me <...>

J.P. Morgan, 1912

Social networks:
a reflection of personal features, achievements, social connections
they reveal what a person really is






Would you trust a man more if you knew about what he really is and not only what he has?

Blended
Scoring

Social scoring
Benefits
Augmenting
decisions
for limited
credit history
Scoring
people
with no
credit history
Has credit
history
No credit
history
Age <=25
Age >25
Information within score
Age & gender
Marital status
Languages
Emigration status
Education & work
...
Profile information
Hobbies & interests
Visited countries
Games & gambling activities
Photos
Groups
iPhone/Android
...
Communication culture
Third-party application posts
Uppercase & special symbols
Swear words
...
Usage level

Status updates
Posts & likes
Time spent in working hours
Subscribers & followers
Social groups
...
Friends & communities
Family and relatives
Friends' age distribution
Structure of communities
...
Photo analysis
Looking deeper for more Volume and Variety
Photos classified by core features - objects in photos
Personal profiles identified - features distribution
Credit risk - significant personal profile differences

"Good" photo
profile
"Bad"
photo
profile
Photo profile
15-20% Gini
Classic score:
Social score:
Demographics
Credit history
Credit activity
Employment info
Linkages with business
...
Profile information
Communication culture
Usage level
Hobbies & interests
Friends & communities
...
Variables examples
Head of Econometric Modelling Department
MSc in Econometrics
10 years of experience in applying econometric modelling and machine learning techniques in credit risk modelling
Mantas Tartėnas
Nataliya Soldatyuk
xxx
yyy
zzz
No credit history or income record -
potential within unbanked people
Thin credit history file -
may not reflect up to date situation
SME's credit scoring:
traditional data + blended scoring
Undeclared electricity
consumption and thievery
1.6 million objects scored (households + businesses)
700 million rows of data - 100 GB+ of data processed
~70% Gini, detection rate increased 500%

Utility provider's data
used for default prediction
SME's credit scoring:
traditional data + blended scoring
Gini ~45% - close to some traditional MFI models
15-20% weight in credit risk models
Credit risk modelling
using social network data
Halfway of presentation
Exponential data growth - increased alternative data opportunities
Alternative data - solutions for clients with no credit history
Big data methods can be used in credit scoring
Social scoring - alternative way of credit scoring
Facebook login
metadata
source of a post
(i.e. Facebook games)
tagged places in photos
Scoring with scorecards
a statistical model that predicts the performance of subjects in the future
table structure:
variables - attributes - scores
Scorecard
Scorecards - standard in financial institutions

Technological advance & exponential data growth -
many unexplored & unusual scorecard opportunities.
Blended scoring - scoring a network of connected business subjects
100k SME's in Lithuania
50% with network connections
320k direct and indirect linkages
(parental or
subsidiary companies)
Direct linkages
Indirect linkages
(same manager,
board member or shareholder)
Topics
Big data scoring - traditional and alternative data
Social network data scoring
Blended scoring
Text mining
Text
mining
Metadata
analysis
Metadata analysis
Photos
Photos
Source : client's portfolio analysis
Note: although correlated, social data is not a substitute for classic data

Gini ~40%
Financier and banker
...
...
...
...
sports, dining, travelling, ...
clubbing, drinking, ...
photo upload intensity
Marketing
Optimizing marketing campaigns by targeting specific people groups
Key points:
Volume
Veracity
Velocity
Variety
V
V
V
V
Volume
Veracity
Velocity
Variety
V
V
V
V
Full transcript