Predictive Data Modeling A CASE STUDY FOR DATA MODELING.

Predictive Data ModelingA CASE STUDY FOR DATA MODELING

•Ken Reed, Loss Control Services Manager

•Barbara Rhoades, Chief Executive Officer

Towers Watson

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Data Analysis

•Why capturing your data is important…

•What you do with it – how to react to it…

How We Have Used Our Data•OTRP manages the process through the Loss Control Summit

•Majority of our risk is Auto Liability

•All of our data is based on what we know…

Annual OTRP Loss Control Summit

Lo

ss C

on

tro

l Su

mm

it 2

01

3

Working Together to

Reduce Transit Losses

Standing Strong

• Provide annually updated analysis of loss trends.

• Use data to drive loss control efforts and employment of resources

• Create awareness of emerging risk/exposures and develop responses

•Create opportunities to share ideas and resources

Loss data from 1/1/2003 to 11/30/2011Event basedOnly looked at events with incurred values

Total of 1910 EventsTotal Incurred of $14,152,647Average Loss per Event of $7,410

Examples of How we have presented our Loss Data

All Events by Year

2003 2004 2005 2006 2007 2008 2009 2010 20110

50

100

150

200

250

300

186

219

267

242235

213

187195

166

All Incurred by Year

2003 2004 2005 2006 2007 2008 2009 2010 2011 $-

$500,000

$1,000,000

$1,500,000

$2,000,000

$2,500,000

$3,000,000

$3,500,000

$3,477,113

$1,514,118

$1,784,810

$2,548,735

$1,321,615

$1,017,440 $879,720 $891,568

$717,529

Average per Event by Year

2003 2004 2005 2006 2007 2008 2009 2010 2011 $-

$2,000

$4,000

$6,000

$8,000

$10,000

$12,000

$14,000

$16,000

$18,000

$20,000

$18,694

$6,914 $6,685

$10,532

$5,624 $4,777 $4,704

$4,572 $4,322

Events by Type

Collision; 1485; 78%

Passenger; 181; 9%

Wheelchair; 84; 4%

Pedestrian; 42; 2%

Misc/Other; 118; 6%

Incurred by Type

Collision7093782.38999

99952%

Passenger955308.179999998

7%

Wheelchair1081283.93

8%

Pedestrian3672274.15

27%

Misc/Other825969.85

6%

Average per Event by Type

Collision

Passenger

Wheelchair

Pedestrian

Misc/Other

$- $10,000 $20,000 $30,000 $40,000 $50,000 $60,000 $70,000 $80,000 $90,000 $100,000

$4,782

$5,278

$12,829

$99,809

$7,001

Average/Event

Collisions Total of 1485 Collision Events Total Incurred of $7,101,609 Average Loss per Collision of $4,782 Collision breakdown

1. Fixed Object2. Rear End Collision3. Backing4. All Other

Collision Events by Year

2003 2004 2005 2006 2007 2008 2009 2010 20110

50

100

150

200

250

129

168

209

181191

161152

167

127

Collision Incurred by Year

2003 2004 2005 2006 2007 2008 2009 2010 2011 $-

$500,000

$1,000,000

$1,500,000

$2,000,000

$2,500,000

$364,068

$691,721 $732,517

$2,300,449

$655,093 $546,126

$609,577 $629,429 $572,630

Collisions by TypeFixed Ob-ject; 300;

20%

Rear End; 262; 18%

Backing; 113; 8%

All Other; 810; 55%

Incurred by Collision TypeFixed Ob-

ject; 919375.09;

13%

Rear End; 1928816.93; 27%

Backing; 155492.41;

2%

All Other; 4097924.69; 58%

Average per Collision by Type

Fixed Object

Rear End

Backing

All Other

$- $1,000 $2,000 $3,000 $4,000 $5,000 $6,000 $7,000 $8,000

$3,065

$7,362

$1,376

$5,059

What is Predictive Modeling?

Historical Actuarial PerspectiveHistorically, actuaries evaluated rates to set pricing and reserves. In a traditional approach they look at characteristic separately

Univariate Analysis

Describes the tradition approach of estimating the impact of one factor without considering others

This is often called a “one-way” analysisContent re-printed with permission from Towers Watson ©2013 all rights reserved

Univariate Analysis

Collision

Age of Vehicle

Number of

Miles

Content re-printed with permission from Towers Watson ©2013 all rights reserved

With the advanced in computing technology, actuaries are able to use complex mathematical equations to estimate the impact of all variables simultaneously. This is called..

•Multivariate AnalysisCollisio

nMiles

Age of Vehicl

e

Content re-printed with permission from Towers Watson ©2013 all rights reserved

•Actuarial Term for the computations underlying a multivariate analysis is:

•GLM or Generalized

Linear Modeling

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Benefits of Predictive Modeling• Explore impact of multiple variable simultaneously while taking into account their relationships or correlations

• This allows for more effective pricing in a competitive market

• Better claims handling though identification of claims drivers

• Reduction of risk though probability analysis

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Benefits of Predictive Modeling• Explore impact of multiple variable simultaneously while taking into account their relationships or correlations

• This allows for more effective pricing in a competitive market

• Better claims handling though identification of claims drivers

• Reduction of risk though probability analysis

Content re-printed with permission from Towers Watson ©2013 all rights reserved

How is Predictive Modeling Being Used?

Can Integrate all aspects of your operations and identifies the value of all customers

• Develop Rules• Credit Analysis• Eval Agents• Target Inspections

Underwriting

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Competitive Advantage

PriceImprove Retention

& Satisfaction

Claims Advantages•Reduce Claim costs•Improve efficiency in claim handling process – shorten cycle

•More Accurate reserving practices•Better claim Experience leads to higher customer satisfaction

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Perform a Claims TriageUse Predictive Modeling to identify complicated claims based on multivariate analysis

• Routed to specialists• Correct Experience for Adjuster• Medial Provider?• Attorney involvement• Return to work• Case management

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Content re-printed with permission from Towers Watson ©2013 all rights reserved

Content re-printed with permission from Towers Watson ©2013 all rights reserved

OTRP Utilized to the study to identify patterns of potential future losses

Predictive Data Analysis allowed us to look at our losses differently. We normally analyze what HAS happened.

This allowed us to see what is likely TO happen.

Predictive Data Analysis

Predictive Data Analysis

Predictive Data Analysis

OTRP’s Experience• Re-confirmed what we already knew to be true

• We were able to identify areas of concerns for specific members (not just the pool as a whole)

• Let us think about claim indicators that we might not have otherwise.

FOR OTHERS:

Great management tool for a “hands-off” organization.

Claims management tool for claims intensive organizations.

Creates the possibility to see things in a new and different way

StandingStrong

OHIO TRANSIT RISK POOLService ● Stability ● Security

Serving Public Transit since 1994