Modeling the Settlement Process for Auto Bodily Injury Liability Claims Richard A. Derrig,...
-
Upload
dominic-bradley -
Category
Documents
-
view
217 -
download
0
Transcript of Modeling the Settlement Process for Auto Bodily Injury Liability Claims Richard A. Derrig,...
Modeling the Settlement Modeling the Settlement Process forProcess for
Auto Bodily Injury Liability Auto Bodily Injury Liability ClaimsClaims
Richard A. Derrig,Richard A. Derrig, President, OPAL Consulting LLCPresident, OPAL Consulting LLCVisiting Scholar, Wharton SchoolVisiting Scholar, Wharton School
University of PennsylvaniaUniversity of Pennsylvania
Greg A. RempalaGreg A. RempalaAssociate Professor, StatisticsAssociate Professor, Statistics
University of LouisvilleUniversity of Louisville
CAS Predictive Modeling SeminarBoston, MA
October 4, 2006
AGENDAAGENDA
Auto BI Liability Claims are Auto BI Liability Claims are negotiated not “paid.” negotiated not “paid.”
What are the key components What are the key components of the settlement amount?of the settlement amount?
What is the role of “pain and What is the role of “pain and suffering” payments?suffering” payments?
What role does fraud and What role does fraud and build-up play?build-up play?
What are the key components What are the key components of the settlement negotiation of the settlement negotiation process itself?process itself?
NEGOTIATIONNEGOTIATION
Liability claims are negotiated Liability claims are negotiated not “paid” by the insurernot “paid” by the insurer
First party claims have payment First party claims have payment regulations both good regulations both good (Cooperation) and bad (Time (Cooperation) and bad (Time Frames for Payment) re fraud.Frames for Payment) re fraud.
Negotiation subject only to bad Negotiation subject only to bad faith and unfair claim practice faith and unfair claim practice regulationsregulations
Two-person game: Adjusters and Two-person game: Adjusters and Claimant/Attorneys, but not Claimant/Attorneys, but not suitable for game theory model.suitable for game theory model.
Example in papers is Auto Bodily Example in papers is Auto Bodily Injury Liability – Mass DataInjury Liability – Mass Data
Table 1Table 1
BI Negotiation Leverage Points
Adjuster Advantages
Adjuster has ability to go to trial
Company has the settlement funds
Attorney, provider, or claimant needs money
Adjuster knows history of prior settlements
Adjuster can delay settlement by investigation
Settlement authorization process in company
Initial Determination of Liability
Table 2Table 2
BI Negotiation Leverage Points
Attorney/Claimant Advantages
Attorney/Claimant can build-up specials
Asymmetric information (Accident, Injury, Treatment)
Attorney/Claimant can fail to cooperate
Attorney has experience with company
Investigation costs the company money
Attorney can allege unfair claim practices (93A)
Adjuster under pressure to close files
NEGOTIATIONNEGOTIATION
Claim Payment Claim Payment ComponentsComponents
Demands and OffersDemands and Offers Time Frames for RoundsTime Frames for Rounds Anchoring and AdjustingAnchoring and Adjusting Offer/Demand RatiosOffer/Demand Ratios SettlementsSettlements Mass BI Data for 1996 AYMass BI Data for 1996 AY Statistical ModelingStatistical Modeling
General DamagesGeneral Damages
Special Damages are Special Damages are Claimant Economic LossesClaimant Economic Losses– Medical BillsMedical Bills– Wage LossWage Loss– Other EconomicOther Economic
General Damages (or Pain General Damages (or Pain and Suffering payments) and Suffering payments) are the Residual of are the Residual of Negotiated Settlement Negotiated Settlement Less SpecialsLess Specials– ““Three Times Specials” is a Three Times Specials” is a
MythMyth
Figure 8-31996 Settlement/Specials Ratio Distribution
0.00%
2.00%
4.00%
6.00%
8.00%
10.00%
12.00%
14.00%
16.00%
18.00%
20.00%
0 to 0.5 0.5 to 1 1 to 1.5 1.5 to 2 2 to 2.5 2.5 to 3 3 to 3.5 3.5 to 4 4 to 4.5 4.5 to 5 5 to 5.5 5.5 to 6 6 to 6.5 6.5 to 7 7 to 7.5 7.5 to 8 8 to 8.5 8.5 to 9 9 to 9.5 9.5 to 10 10 to 20 20 to 30
Settlement/Specials Ratio
% of
Claim
s
BI 1996 Negotiations1st and 2nd Demands
$-
$5,000
$10,000
$15,000
$20,000
$25,000
$30,000
$35,000
$40,000
ALL Not in Suit In Suit
Dollar
s
0
50
100
150
200
250
300
350
Claim
Cou
nts
Mean Demand 1
Mean Demand 2
Mean BISettlement
Claim Count
CSE: First & Second Demand Ratio to BI Settlement Ratio
Limited to 2nd Demand > $0, (315 BI Claims)NO PIP payment in Demand & Settlement, Outlier removed 3860
y = 1.4088x + 0.3452
R2 = 0.5691
y = 2.6414x + 1.4777
R2 = 0.1953
0
2
4
6
8
10
12
14
16
18
20
0 2 4 6 8 10 12 14 16 18 20
BI Settlement Ratio
First
& Se
cond
Dem
and R
atio
First Demand
Second Demand
2nd Demand Ratio
1st Demand Ratio
BI Settlement Ratio 1:1
Negotiated Negotiated SettlementsSettlements
Specials may be Specials may be Discounted or IgnoredDiscounted or Ignored
Medicals: Real or Built-up?Medicals: Real or Built-up? Information from Information from
InvestigationInvestigation Independent Medical Independent Medical
Exams (IMEs)Exams (IMEs) Special InvestigationSpecial Investigation Suspicion of Fraud or Build-Suspicion of Fraud or Build-
upup
Settlement Ratios by Injury and SuspicionSettlement Ratios by Injury and SuspicionVariableVariable PIP Suspicion PIP Suspicion
Score Score
= Low (0-3)= Low (0-3)
PIP Suspicion PIP Suspicion Score Score
= Mod to = Mod to High (4-10)High (4-10)
PIP PIP Suspicion Suspicion
Score = AllScore = All
1996 (N-336)1996 (N-336) 1996 (N-216)1996 (N-216) 1996 (N-1996 (N-552)552)
Str/Str/SPSP
All All OtherOther
Str/SPStr/SP All All OtherOther
Str/Str/SPSP
All All OtheOtherr
SettlementSettlement SettlementSettlement SettlementSettlement
81%81% 19%19% 94%94% 6%6% 86%86% 14%14%
Avg. Avg. Settlement/Settlement/Specials Specials RatioRatio
3.013.01 3.813.81 2.582.58 3.613.61 2.822.82 3.773.77
Median Median Settlement/Settlement/Specials Specials RatioRatio
2.692.69 2.892.89 2.402.40 2.572.57 2.552.55 2.892.89
Settlement ModelingSettlement Modeling
Major Claim Major Claim CharacteristicsCharacteristics
Tobit Regression for Tobit Regression for Censored Data (right Censored Data (right censored for policy limits)censored for policy limits)
Evaluation Model for Evaluation Model for Objective “Facts”Objective “Facts”
Negotiation Model for all Negotiation Model for all Other “Facts”, including Other “Facts”, including suspicion of fraud or suspicion of fraud or build-up build-up
Evaluation VariablesEvaluation Variables
Tobit Model (1996AYTobit Model (1996AY)) Claimed Medicals (+)Claimed Medicals (+) Claimed Wages (+)Claimed Wages (+) Fault (+)Fault (+) Attorney (+18%)Attorney (+18%) Fracture (+82%)Fracture (+82%) Serious Visible Injury Scene Serious Visible Injury Scene
(+36%)(+36%) Disability Weeks (+10% @ 3 Disability Weeks (+10% @ 3
weeks)weeks) Non-Emergency CT/MRI (+31%)Non-Emergency CT/MRI (+31%) Low Impact Collision (-14%)Low Impact Collision (-14%) Three Claimants in Vehicle (-Three Claimants in Vehicle (-
12%)12%) Same BI + PIP Co. (-10%) Same BI + PIP Co. (-10%)
[Passengers -22%] [Passengers -22%]
Negotiation VariablesNegotiation Variables Model Additions (1996AY)Model Additions (1996AY)
Atty (1st) Demand Ratio to Specials Atty (1st) Demand Ratio to Specials (+8% @ 6 X Specials)(+8% @ 6 X Specials)
BI IME No Show (-30%) BI IME No Show (-30%) BI IME Positive Outcome (-15%)BI IME Positive Outcome (-15%) BI IME Not Requested (-14%)BI IME Not Requested (-14%) BI Ten Point Suspicion Score (-12% @ BI Ten Point Suspicion Score (-12% @
5.0 Average)5.0 Average) [1993 Build-up Variable (-10%)][1993 Build-up Variable (-10%)] Unknown Disability (+53%)Unknown Disability (+53%) [93A (Bad Faith) Letter Not Significant][93A (Bad Faith) Letter Not Significant] [In Suit Not Significant][In Suit Not Significant] [SIU Referral (-6%) but Not Significant][SIU Referral (-6%) but Not Significant] [EUO Not Significant][EUO Not Significant]
Note: PIP IME No Show also significantly Note: PIP IME No Show also significantly reduces BI + PIP by discouraging BI reduces BI + PIP by discouraging BI claim altogether (-3%).claim altogether (-3%).
Total Value of Negotiation Total Value of Negotiation VariablesVariables
Total Compensation Total Compensation VariablesVariables
Avg. Avg. Claim/FactorClaim/Factor
Evaluation VariablesEvaluation Variables $13,948$13,948
Disability UnknownDisability Unknown 1.051.05
11stst Demand Ratio Demand Ratio 1.091.09
BI IME No ShowBI IME No Show 0.990.99
BI IME Not RequestedBI IME Not Requested 0.900.90
BI IME Performed with BI IME Performed with Positive OutcomePositive Outcome
0.970.97
SuspicionSuspicion 0.870.87
Negotiation VariablesNegotiation Variables 0.870.87
Total Compensation Model Total Compensation Model PaymentPayment
$12,058$12,058
Actual Total CompensationActual Total Compensation $11,863$11,863
Actual BI PaymentActual BI Payment $8,551$8,551
Actual parameters Actual parameters for negotiation and for negotiation and evaluation models, evaluation models, with and without with and without
suspicion variable, suspicion variable, are shown in the are shown in the
hard copy handouthard copy handout
NEGOTIATIONNEGOTIATION
Claim Payment Claim Payment ComponentsComponents
Demands and OffersDemands and Offers Time Frames for RoundsTime Frames for Rounds Anchoring and AdjustingAnchoring and Adjusting Offer/Demand RatiosOffer/Demand Ratios SettlementsSettlements Mass BI Data for 1996 AYMass BI Data for 1996 AY Statistical ModelingStatistical Modeling
STAT. MODELINGSTAT. MODELING
Identify Identify random componentrandom component of of negotiation process (in any)negotiation process (in any)
Demands and offers not Demands and offers not independent independent
Claims sizes form mixtures of Claims sizes form mixtures of dists dists
Assume: current O (D) depend Assume: current O (D) depend only on the previous O, Donly on the previous O, D
Markov Chain ?Markov Chain ? Time frames for rounds seem Time frames for rounds seem
homonegous (possibly homonegous (possibly deterministic) deterministic)
Consider O/D values in a single Consider O/D values in a single claim negotiation claim negotiation
A Statistical Analysis of the A Statistical Analysis of the Effect of Anchoring in the Effect of Anchoring in the
Negotiation Process of Negotiation Process of Automobile Bodily Injury Automobile Bodily Injury
Liability ClaimsLiability Claims
Richard A. Derrig,Richard A. Derrig, President, OPAL Consulting LLCPresident, OPAL Consulting LLCVisiting Scholar, Wharton SchoolVisiting Scholar, Wharton School
University of PennsylvaniaUniversity of Pennsylvania
Greg A. RempalaGreg A. RempalaAssociate Professor, StatisticsAssociate Professor, Statistics
University of LouisvilleUniversity of Louisville
Working Paper v 3.1Working Paper v 3.1March 10, 2006March 10, 2006
Table 6Table 6
Negotiation – Offer/Demand Ratios by Round
4 ROUNDS (100 claims)
O1/D1 O2/D2 O3/D3 BI/D3
Average 0.246 0.476 0.724 0.798
Std. Dev. 0.153 0.213 0.211 0.191
3 ROUNDS (119 claims)
O1/D1 O2/D2 BI/D2
Average 0.393 0.708 0.766
Std. Dev. 0.610 0.212 0.191
O/D ProcessO/D Process
0 1
Initial Settlement
O1/D1 O2/D2 O3/D3
OOii/D/Dii values are non values are non decreasing, should tend to decreasing, should tend to one (settlement)one (settlement) Considering O/D Considering O/D homogenizes the data from homogenizes the data from different claim negotiations, different claim negotiations, but:but: Disregards Disregards timetime and and claim claim sizesize Possibly removes some Possibly removes some other covariates (Injury, etc)other covariates (Injury, etc)
Offer Demand Ratios (Sorted Offer Demand Ratios (Sorted by Descending Losses) – by Descending Losses) –
FigureFigure 11
Offer Demand Ratios (Sorted by Offer Demand Ratios (Sorted by Descending 1st Demands) – Figure Descending 1st Demands) – Figure
22
O/D as Poisson ProcessO/D as Poisson Process
Nt number of discrete events on (0,t] arriving “one at a time”
Nt is NHPP with rate (t), if for every t>0
P(Nt =k)=exp(-z(t)) [z(t)] k/k!.
where z(t)=0t (s)ds
NHPP is uniquely determined by its rate function (t)
Distance between Oi/Di and Oi+1/Di+1 is exponential with rate (t)
How to estimate (t) ?
Rate EstimationRate Estimation
(t) may be approximated by a piecewise function
Decide on a time interval within Decide on a time interval within which rate is fixedwhich rate is fixed
Estimate from O/D data the Estimate from O/D data the (constant) rate during each (constant) rate during each intervalinterval
Easy simulation of NHPP with Easy simulation of NHPP with piecewise constant piecewise constant (t) using rejection method
t
( ) t
Rates ComparisonRates Comparison
(t) is the average “speed” of negotiation measured in O/D ratio increase rate
Is it the same for all claims ? Simple statistical test based on
parametric resampling 95 % confidence envelopes
(tunnels) No evidence of difference in
(t) for 3 and 4 rounds (lay within each other tunnels)
(t) for 2 round is significantly different
Figure 1: Figure 1: The Massachusetts Negotiation Data The Massachusetts Negotiation Data
Estimated standardized rates of the NHPP of Estimated standardized rates of the NHPP of arrival of O/D for 2-, 3- and 4-negotiation arrival of O/D for 2-, 3- and 4-negotiation
rounds.rounds.
Rates comparison (cont)Rates comparison (cont)
Seems that the Mass. data Seems that the Mass. data induces two types of rates:induces two types of rates:
Slow rate (2 rounds)Slow rate (2 rounds) Fast rate (3 or more rounds) Fast rate (3 or more rounds) Can we predict the rate type Can we predict the rate type
from the initial set of covariates from the initial set of covariates ??
Use logistic regression for Use logistic regression for classificationclassification
Simple, yet satisfying (error: Simple, yet satisfying (error: 18% on data, 20% on cross-18% on data, 20% on cross-valildation)valildation)
Comparable to SVM and othersComparable to SVM and others
Table 10Table 10
Logistic Classifier of Fast and Slow Claims
Variable CoefficientStandard
Error p-Value
Demand 1 (000's) -0.0678 0.0327 0.0385
O1 / D1 -5.4660 2.8440 0.0546
Report Date – Accident Date (days) -0.0297 0.0103 0.0038
Three or more claimants -1.6990 1.0580 0.1082
BI IME Not Requested 3.1300 1.0940 0.0042
BI IME Performed with Positive Outcome 2.5460 1.4490 0.0789
Intercept 3.0120 1.7000
0.0764
Figure 3:Figure 3:95% confidence tunnel for both ‘slow’ and 95% confidence tunnel for both ‘slow’ and
‘fast’ fitted rates for the subset of 58 ‘fast’ fitted rates for the subset of 58 negotiations histories from the negotiations histories from the
Massachusetts datasetMassachusetts dataset
Table 7Table 7
Offer/Demand Ratio Dependence on Demand
Ratio Rounds Intercept Int.S.E.
Demand (000)
Coefficient
O1/D1 2 0.55 0.08 -0.0074
O2/D2 2 0.78 0.02 -0.0061
BI/D2 2 0.84 0.02 -0.0057
O1/D1 3 0.28 0.02 -0.0013
O2/D2 3 0.56 0.03 -0.0055
O3/D3 3 0.79 0.03 -0.0058
BI/D3 3 0.84 0.02 -0.0040
All intercept and demand coefficients significant at 1%
Offer / Demand Ratios Offer / Demand Ratios (Sorted by Descending Pre-(Sorted by Descending Pre-Settlement Ratio) – Figure 3Settlement Ratio) – Figure 3
Simulated vs True O/D DataSimulated vs True O/D Data
Alternative approach:Alternative approach:SVM classifier SVM classifier
Drive a hyperplane across data to Drive a hyperplane across data to separate FAST/SLOW claims separate FAST/SLOW claims Prediction: On which side of the hyperplane does the new point lie? Points in the direction of the normal vector are classified as POSITIVE (fast); otherwise NEGATIVE (slow).
Alternative approach:Alternative approach:SVM classifier (cont)SVM classifier (cont)
If data separable, pick a hyperplane with If data separable, pick a hyperplane with largest possible margin largest possible margin Otherwise penalty for misclassification Often Often data may be separable after space transformation
NEGOTIATIONNEGOTIATIONFuture Modeling Future Modeling
Work Work Demands and OffersDemands and Offers Role of Time Frames Role of Time Frames Role of Covariates (Injury, Role of Covariates (Injury,
etc)etc) Anchoring and AdjustingAnchoring and Adjusting Offer/Demand RatiosOffer/Demand Ratios SettlementsSettlements Statistical ModelsStatistical Models Mass BI Data for 1996 AYMass BI Data for 1996 AY Another Data Set NeededAnother Data Set Needed
ReferencesReferences Cooter, Robert D. and Daniel L. Rubinfeld, (1989), Cooter, Robert D. and Daniel L. Rubinfeld, (1989),
Economic Analysis of Legal Disputes and Their Economic Analysis of Legal Disputes and Their Resolution, Resolution, Journal of Economic LiteratureJournal of Economic Literature, 27, 1067-, 27, 1067-10971097
Derrig, Richard, and Herbert I. Weisberg, (2004), Derrig, Richard, and Herbert I. Weisberg, (2004), Determinants of Total Compensation for Auto Bodily Determinants of Total Compensation for Auto Bodily Injury Liability Under No Fault: Investigation, Injury Liability Under No Fault: Investigation, Negotiation and the Suspicion of Fraud, Negotiation and the Suspicion of Fraud, Insurance and Insurance and Risk ManagementRisk Management, 71:4, 633-662, January., 71:4, 633-662, January.
Epley, Nicholas, and Thomas Gilovich, (2001), Putting Epley, Nicholas, and Thomas Gilovich, (2001), Putting Adjustment Back in the Anchoring and Adjustment Adjustment Back in the Anchoring and Adjustment Heuristic: Differential Processing of Self-Generated and Heuristic: Differential Processing of Self-Generated and Experimenter-Provided Anchors, Experimenter-Provided Anchors, Psychological SciencePsychological Science, , 12:5, 391-396.12:5, 391-396.
Loughran, David, (2005) Deterring Fraud: The Role of Loughran, David, (2005) Deterring Fraud: The Role of General Damage Awards in Automobile Insurance General Damage Awards in Automobile Insurance Settlements, Settlements, Journal of Risk and Insurance, Journal of Risk and Insurance, 72:551-57572:551-575
Raiffa, Howard, (1982), Raiffa, Howard, (1982), The Art and Science of The Art and Science of Negotiation, Negotiation, The Belknap Press of Harvard University The Belknap Press of Harvard University Press.Press.
Ross, Lawrence, H., (1980), Ross, Lawrence, H., (1980), Settled Out of CourtSettled Out of Court, , (Chicago, III: Aldine).(Chicago, III: Aldine).
Tversky, A., and D. Kahneman, (1974), Judgment Under Tversky, A., and D. Kahneman, (1974), Judgment Under Uncertainty: Heuristics and Biases, Uncertainty: Heuristics and Biases, ScienceScience, 195, 1124-, 195, 1124-1130.1130.
Wright, W.F. and U. Anderson, (1989), Effects of Wright, W.F. and U. Anderson, (1989), Effects of Situation Familiarity and Incentives on use of the Situation Familiarity and Incentives on use of the Anchoring and Adjustment Heuristic for Probability Anchoring and Adjustment Heuristic for Probability Assessment, Assessment, Organizational Behavior and Human Organizational Behavior and Human Decision ProcessesDecision Processes, 44, 68-82., 44, 68-82.