Download presentation
Presentation is loading. Please wait.
Published byCornelius Heath Modified over 8 years ago
1
Adrienne Lewis & Kyle Hales November 21, 2013 1
2
Definition The process by which a model is created to try to best predict the probability of an outcome Limitations Data – delicate balance between homogeneity and credibility or potential gaps History cannot always predict the future – model implicitly assumes certain steady-state conditions which usually don’t hold when people are involved Issue of unknown unknowns – undiscovered variables may be critical to successful model Manipulation by others 2
3
Competitive Intelligence Research peers in line and/or key state(s) May provide starting place for variable investigation Model Building GLM or tiering approach Internal company data or external data Single company or multi-company Filing Support Actuarial justification Create rules to explain model rating steps Keep proprietary models as confidential as state allows 3
4
Lines of Business & Model Types Industry News External Data 4
5
Commercial Auto Rating Businessowners (BOP) Rating Workers Compensation (WC) Rating/Tiering Underwriting Claims General Liability (GL) Tiering 5
6
Generalized Linear Models (GLMs) ◦ Continuous ◦ Random component ◦ Systematic component ◦ Link function ◦ Complex – five or more elements Tiering Criteria ◦ Discrete ◦ Simple – three to five elements 6
7
In order to focus underwriting on the good risks, insurers need a model to determine Potential variables: Physician data (area of specialty, location, billing/payment history, practice demographics) Claims data (experience data, frequency, loss control, fraud) Third-party databases (MVRs, credit, AMA master file, pharmacy data) "It's the Right Time for Right Pricing in Medical Malpractice Insurance“ Contingencies, July/August 2005 7
8
Provides the best match for predictive modeling due to the data intensity already required for reporting Uses of predictive modeling for workers compensation: Prospective identification of high cost claims (reserving, case management, settlement strategy) Premium auditing (when and whom to audit) Fraud detection (earlier and more accurately) "Workers' Comp and Predictive Modeling" The CLM, May 2012 8
9
Valen Technologies Inc. Develops P&C predictive models Interest increased in the past 18 months Insurers may look to predictive modeling for a competitive edge in the difficult workers compensation market. CNA Created a predictive model and determined which factors make some companies safer than others Use model in underwriting to predict accidents and determine the quality of an account Model helps CNA recommend safety measures to insureds to reduce workers compensation losses "Predictive Models Increasingly Being Used as Workers Comp Underwriting Tool" Business Insurance, February 2013 9
10
Bayesian belief based models – model the decision making process of expert underwriters Design and test with minor investment of time and cost; adjusts over time Comprised of three elements: Risk appetite gate – only submissions in the insurer’s risk appetite definitions are evaluated by its underwriters Risk quality predictive models – one model for each product line to output a score Bayesian predictive methodology for package submissions – takes scores from each product line to assess the overall score of the account "Using Predictive Modeling to Improve Underwriting Performance for Middle-Market Commercial Insurance Accounts" TNC Management Group, March 2009 10
11
LexisNexis ◦ Commercial Attract Score Commercial credit data (Dun & Bradstreet and Experian) Business demographics Business owner data (including personal credit & claims) Woods & Poole Economics, Inc. ◦ Database contains 900 economic and demographic variables by state, region, county, metro area Population data by age, sex and race Employment and earnings by major industry Retail sales by kind of business 1970 to 2040 11
12
Pitney Bowes Business demographics Includes SIC/NAICS, employee count, sales volume ISO/Verisk Environmental Module - Commercial Auto & BOP Workers Compensation premium audit model Social Intelligence Corp ◦ Search social media To properly classify or uncover aspects of businesses missed by traditional underwriting To expedite and aid in fraud detection 12
13
Marshall & Swift/Boeckh PerilVision – property losses by peril BVS Commercial – rebuild costs by business type Hazard data – natural zones (flood, sink hole, wind, etc) RiskMeter Online Property location data Crime report CoreLogic Natural hazard risk Commercial market trends (sales, loan originations, defaults) by segment 13
14
Frequency Severity Loss Ratio Frequency Severity Loss Ratio Quotes Sales Hit Ratio Quotes Sales Hit Ratio Credit Location Crime Credit Location Crime Competition Demographics Opportunity Competition Demographics Opportunity 14
15
State Matrix Examples 15
16
16
17
Stored on Sharepoint Company Filings 14 reference filings (Auto, BOP, GL, WC) If you encounter a filing with a predictive model, save it to Sharepoint for others to reference Articles Predictive modeling surveys (Towers Watson, ISO) Presentations This document Filing Support Matrix for Commercial Lines Technical Papers 17
18
18
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.